/* * HRTF utility for producing and demonstrating the process of creating an * OpenAL Soft compatible HRIR data set. * * Copyright (C) 2011-2019 Christopher Fitzgerald * * This program is free software; you can redistribute it and/or modify * it under the terms of the GNU General Public License as published by * the Free Software Foundation; either version 2 of the License, or * (at your option) any later version. * * This program is distributed in the hope that it will be useful, * but WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the * GNU General Public License for more details. * * You should have received a copy of the GNU General Public License along * with this program; if not, write to the Free Software Foundation, Inc., * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. * * Or visit: http://www.gnu.org/licenses/old-licenses/gpl-2.0.html * * -------------------------------------------------------------------------- * * A big thanks goes out to all those whose work done in the field of * binaural sound synthesis using measured HRTFs makes this utility and the * OpenAL Soft implementation possible. * * The algorithm for diffuse-field equalization was adapted from the work * done by Rio Emmanuel and Larcher Veronique of IRCAM and Bill Gardner of * MIT Media Laboratory. It operates as follows: * * 1. Take the FFT of each HRIR and only keep the magnitude responses. * 2. Calculate the diffuse-field power-average of all HRIRs weighted by * their contribution to the total surface area covered by their * measurement. This has since been modified to use coverage volume for * multi-field HRIR data sets. * 3. Take the diffuse-field average and limit its magnitude range. * 4. Equalize the responses by using the inverse of the diffuse-field * average. * 5. Reconstruct the minimum-phase responses. * 5. Zero the DC component. * 6. IFFT the result and truncate to the desired-length minimum-phase FIR. * * The spherical head algorithm for calculating propagation delay was adapted * from the paper: * * Modeling Interaural Time Difference Assuming a Spherical Head * Joel David Miller * Music 150, Musical Acoustics, Stanford University * December 2, 2001 * * The formulae for calculating the Kaiser window metrics are from the * the textbook: * * Discrete-Time Signal Processing * Alan V. Oppenheim and Ronald W. Schafer * Prentice-Hall Signal Processing Series * 1999 */ #include "config.h" #define _UNICODE #include #include #include #include #include #include #include #include #include #ifdef HAVE_STRINGS_H #include #endif #ifdef HAVE_GETOPT #include #else #include "getopt.h" #endif #include #include #include #include #include #include #include #include #include #include "mysofa.h" #include "win_main_utf8.h" namespace { using namespace std::placeholders; } // namespace #ifndef M_PI #define M_PI (3.14159265358979323846) #endif // The epsilon used to maintain signal stability. #define EPSILON (1e-9) // Constants for accessing the token reader's ring buffer. #define TR_RING_BITS (16) #define TR_RING_SIZE (1 << TR_RING_BITS) #define TR_RING_MASK (TR_RING_SIZE - 1) // The token reader's load interval in bytes. #define TR_LOAD_SIZE (TR_RING_SIZE >> 2) // The maximum identifier length used when processing the data set // definition. #define MAX_IDENT_LEN (16) // The maximum path length used when processing filenames. #define MAX_PATH_LEN (256) // The limits for the sample 'rate' metric in the data set definition and for // resampling. #define MIN_RATE (32000) #define MAX_RATE (96000) // The limits for the HRIR 'points' metric in the data set definition. #define MIN_POINTS (16) #define MAX_POINTS (8192) // The limit to the number of 'distances' listed in the data set definition. #define MAX_FD_COUNT (16) // The limits to the number of 'azimuths' listed in the data set definition. #define MIN_EV_COUNT (5) #define MAX_EV_COUNT (128) // The limits for each of the 'azimuths' listed in the data set definition. #define MIN_AZ_COUNT (1) #define MAX_AZ_COUNT (128) // The limits for the listener's head 'radius' in the data set definition. #define MIN_RADIUS (0.05) #define MAX_RADIUS (0.15) // The limits for the 'distance' from source to listener for each field in // the definition file. #define MIN_DISTANCE (0.05) #define MAX_DISTANCE (2.50) // The maximum number of channels that can be addressed for a WAVE file // source listed in the data set definition. #define MAX_WAVE_CHANNELS (65535) // The limits to the byte size for a binary source listed in the definition // file. #define MIN_BIN_SIZE (2) #define MAX_BIN_SIZE (4) // The minimum number of significant bits for binary sources listed in the // data set definition. The maximum is calculated from the byte size. #define MIN_BIN_BITS (16) // The limits to the number of significant bits for an ASCII source listed in // the data set definition. #define MIN_ASCII_BITS (16) #define MAX_ASCII_BITS (32) // The limits to the FFT window size override on the command line. #define MIN_FFTSIZE (65536) #define MAX_FFTSIZE (131072) // The limits to the equalization range limit on the command line. #define MIN_LIMIT (2.0) #define MAX_LIMIT (120.0) // The limits to the truncation window size on the command line. #define MIN_TRUNCSIZE (16) #define MAX_TRUNCSIZE (512) // The limits to the custom head radius on the command line. #define MIN_CUSTOM_RADIUS (0.05) #define MAX_CUSTOM_RADIUS (0.15) // The truncation window size must be a multiple of the below value to allow // for vectorized convolution. #define MOD_TRUNCSIZE (8) // The defaults for the command line options. #define DEFAULT_FFTSIZE (65536) #define DEFAULT_EQUALIZE (1) #define DEFAULT_SURFACE (1) #define DEFAULT_LIMIT (24.0) #define DEFAULT_TRUNCSIZE (32) #define DEFAULT_HEAD_MODEL (HM_DATASET) #define DEFAULT_CUSTOM_RADIUS (0.0) // The four-character-codes for RIFF/RIFX WAVE file chunks. #define FOURCC_RIFF (0x46464952) // 'RIFF' #define FOURCC_RIFX (0x58464952) // 'RIFX' #define FOURCC_WAVE (0x45564157) // 'WAVE' #define FOURCC_FMT (0x20746D66) // 'fmt ' #define FOURCC_DATA (0x61746164) // 'data' #define FOURCC_LIST (0x5453494C) // 'LIST' #define FOURCC_WAVL (0x6C766177) // 'wavl' #define FOURCC_SLNT (0x746E6C73) // 'slnt' // The supported wave formats. #define WAVE_FORMAT_PCM (0x0001) #define WAVE_FORMAT_IEEE_FLOAT (0x0003) #define WAVE_FORMAT_EXTENSIBLE (0xFFFE) // The maximum propagation delay value supported by OpenAL Soft. #define MAX_HRTD (63.0) // The OpenAL Soft HRTF format marker. It stands for minimum-phase head // response protocol 02. #define MHR_FORMAT ("MinPHR02") // Sample and channel type enum values. enum SampleTypeT { ST_S16 = 0, ST_S24 = 1 }; // Certain iterations rely on these integer enum values. enum ChannelTypeT { CT_NONE = -1, CT_MONO = 0, CT_STEREO = 1 }; // Byte order for the serialization routines. enum ByteOrderT { BO_NONE, BO_LITTLE, BO_BIG }; // Source format for the references listed in the data set definition. enum SourceFormatT { SF_NONE, SF_ASCII, // ASCII text file. SF_BIN_LE, // Little-endian binary file. SF_BIN_BE, // Big-endian binary file. SF_WAVE, // RIFF/RIFX WAVE file. SF_SOFA // Spatially Oriented Format for Accoustics (SOFA) file. }; // Element types for the references listed in the data set definition. enum ElementTypeT { ET_NONE, ET_INT, // Integer elements. ET_FP // Floating-point elements. }; // Head model used for calculating the impulse delays. enum HeadModelT { HM_NONE, HM_DATASET, // Measure the onset from the dataset. HM_SPHERE // Calculate the onset using a spherical head model. }; /* Unsigned integer type. */ using uint = unsigned int; /* Complex double type. */ using complex_d = std::complex; /* Channel index enums. Mono uses LeftChannel only. */ enum ChannelIndex : uint { LeftChannel = 0u, RightChannel = 1u }; // Token reader state for parsing the data set definition. struct TokenReaderT { FILE *mFile; const char *mName; uint mLine; uint mColumn; char mRing[TR_RING_SIZE]; size_t mIn; size_t mOut; }; // Source reference state used when loading sources. struct SourceRefT { SourceFormatT mFormat; ElementTypeT mType; uint mSize; int mBits; uint mChannel; double mAzimuth; double mElevation; double mRadius; uint mSkip; uint mOffset; char mPath[MAX_PATH_LEN+1]; }; // Structured HRIR storage for stereo azimuth pairs, elevations, and fields. struct HrirAzT { double mAzimuth{0.0}; uint mIndex{0u}; double mDelays[2]{0.0, 0.0}; double *mIrs[2]{nullptr, nullptr}; }; struct HrirEvT { double mElevation{0.0}; uint mIrCount{0u}; uint mAzCount{0u}; HrirAzT *mAzs{nullptr}; }; struct HrirFdT { double mDistance{0.0}; uint mIrCount{0u}; uint mEvCount{0u}; uint mEvStart{0u}; HrirEvT *mEvs{nullptr}; }; // The HRIR metrics and data set used when loading, processing, and storing // the resulting HRTF. struct HrirDataT { uint mIrRate{0u}; SampleTypeT mSampleType{ST_S24}; ChannelTypeT mChannelType{CT_NONE}; uint mIrPoints{0u}; uint mFftSize{0u}; uint mIrSize{0u}; double mRadius{0.0}; uint mIrCount{0u}; uint mFdCount{0u}; std::vector mHrirsBase; std::vector mEvsBase; std::vector mAzsBase; std::vector mFds; }; // The resampler metrics and FIR filter. struct ResamplerT { uint mP, mQ, mM, mL; std::vector mF; }; /***************************** *** Token reader routines *** *****************************/ /* Whitespace is not significant. It can process tokens as identifiers, numbers * (integer and floating-point), strings, and operators. Strings must be * encapsulated by double-quotes and cannot span multiple lines. */ // Setup the reader on the given file. The filename can be NULL if no error // output is desired. static void TrSetup(FILE *fp, const char *filename, TokenReaderT *tr) { const char *name = nullptr; if(filename) { const char *slash = strrchr(filename, '/'); if(slash) { const char *bslash = strrchr(slash+1, '\\'); if(bslash) name = bslash+1; else name = slash+1; } else { const char *bslash = strrchr(filename, '\\'); if(bslash) name = bslash+1; else name = filename; } } tr->mFile = fp; tr->mName = name; tr->mLine = 1; tr->mColumn = 1; tr->mIn = 0; tr->mOut = 0; } // Prime the reader's ring buffer, and return a result indicating that there // is text to process. static int TrLoad(TokenReaderT *tr) { size_t toLoad, in, count; toLoad = TR_RING_SIZE - (tr->mIn - tr->mOut); if(toLoad >= TR_LOAD_SIZE && !feof(tr->mFile)) { // Load TR_LOAD_SIZE (or less if at the end of the file) per read. toLoad = TR_LOAD_SIZE; in = tr->mIn&TR_RING_MASK; count = TR_RING_SIZE - in; if(count < toLoad) { tr->mIn += fread(&tr->mRing[in], 1, count, tr->mFile); tr->mIn += fread(&tr->mRing[0], 1, toLoad-count, tr->mFile); } else tr->mIn += fread(&tr->mRing[in], 1, toLoad, tr->mFile); if(tr->mOut >= TR_RING_SIZE) { tr->mOut -= TR_RING_SIZE; tr->mIn -= TR_RING_SIZE; } } if(tr->mIn > tr->mOut) return 1; return 0; } // Error display routine. Only displays when the base name is not NULL. static void TrErrorVA(const TokenReaderT *tr, uint line, uint column, const char *format, va_list argPtr) { if(!tr->mName) return; fprintf(stderr, "\nError (%s:%u:%u): ", tr->mName, line, column); vfprintf(stderr, format, argPtr); } // Used to display an error at a saved line/column. static void TrErrorAt(const TokenReaderT *tr, uint line, uint column, const char *format, ...) { va_list argPtr; va_start(argPtr, format); TrErrorVA(tr, line, column, format, argPtr); va_end(argPtr); } // Used to display an error at the current line/column. static void TrError(const TokenReaderT *tr, const char *format, ...) { va_list argPtr; va_start(argPtr, format); TrErrorVA(tr, tr->mLine, tr->mColumn, format, argPtr); va_end(argPtr); } // Skips to the next line. static void TrSkipLine(TokenReaderT *tr) { char ch; while(TrLoad(tr)) { ch = tr->mRing[tr->mOut&TR_RING_MASK]; tr->mOut++; if(ch == '\n') { tr->mLine++; tr->mColumn = 1; break; } tr->mColumn ++; } } // Skips to the next token. static int TrSkipWhitespace(TokenReaderT *tr) { while(TrLoad(tr)) { char ch{tr->mRing[tr->mOut&TR_RING_MASK]}; if(isspace(ch)) { tr->mOut++; if(ch == '\n') { tr->mLine++; tr->mColumn = 1; } else tr->mColumn++; } else if(ch == '#') TrSkipLine(tr); else return 1; } return 0; } // Get the line and/or column of the next token (or the end of input). static void TrIndication(TokenReaderT *tr, uint *line, uint *column) { TrSkipWhitespace(tr); if(line) *line = tr->mLine; if(column) *column = tr->mColumn; } // Checks to see if a token is (likely to be) an identifier. It does not // display any errors and will not proceed to the next token. static int TrIsIdent(TokenReaderT *tr) { if(!TrSkipWhitespace(tr)) return 0; char ch{tr->mRing[tr->mOut&TR_RING_MASK]}; return ch == '_' || isalpha(ch); } // Checks to see if a token is the given operator. It does not display any // errors and will not proceed to the next token. static int TrIsOperator(TokenReaderT *tr, const char *op) { size_t out, len; char ch; if(!TrSkipWhitespace(tr)) return 0; out = tr->mOut; len = 0; while(op[len] != '\0' && out < tr->mIn) { ch = tr->mRing[out&TR_RING_MASK]; if(ch != op[len]) break; len++; out++; } if(op[len] == '\0') return 1; return 0; } /* The TrRead*() routines obtain the value of a matching token type. They * display type, form, and boundary errors and will proceed to the next * token. */ // Reads and validates an identifier token. static int TrReadIdent(TokenReaderT *tr, const uint maxLen, char *ident) { uint col, len; char ch; col = tr->mColumn; if(TrSkipWhitespace(tr)) { col = tr->mColumn; ch = tr->mRing[tr->mOut&TR_RING_MASK]; if(ch == '_' || isalpha(ch)) { len = 0; do { if(len < maxLen) ident[len] = ch; len++; tr->mOut++; if(!TrLoad(tr)) break; ch = tr->mRing[tr->mOut&TR_RING_MASK]; } while(ch == '_' || isdigit(ch) || isalpha(ch)); tr->mColumn += len; if(len < maxLen) { ident[len] = '\0'; return 1; } TrErrorAt(tr, tr->mLine, col, "Identifier is too long.\n"); return 0; } } TrErrorAt(tr, tr->mLine, col, "Expected an identifier.\n"); return 0; } // Reads and validates (including bounds) an integer token. static int TrReadInt(TokenReaderT *tr, const int loBound, const int hiBound, int *value) { uint col, digis, len; char ch, temp[64+1]; col = tr->mColumn; if(TrSkipWhitespace(tr)) { col = tr->mColumn; len = 0; ch = tr->mRing[tr->mOut&TR_RING_MASK]; if(ch == '+' || ch == '-') { temp[len] = ch; len++; tr->mOut++; } digis = 0; while(TrLoad(tr)) { ch = tr->mRing[tr->mOut&TR_RING_MASK]; if(!isdigit(ch)) break; if(len < 64) temp[len] = ch; len++; digis++; tr->mOut++; } tr->mColumn += len; if(digis > 0 && ch != '.' && !isalpha(ch)) { if(len > 64) { TrErrorAt(tr, tr->mLine, col, "Integer is too long."); return 0; } temp[len] = '\0'; *value = strtol(temp, nullptr, 10); if(*value < loBound || *value > hiBound) { TrErrorAt(tr, tr->mLine, col, "Expected a value from %d to %d.\n", loBound, hiBound); return 0; } return 1; } } TrErrorAt(tr, tr->mLine, col, "Expected an integer.\n"); return 0; } // Reads and validates (including bounds) a float token. static int TrReadFloat(TokenReaderT *tr, const double loBound, const double hiBound, double *value) { uint col, digis, len; char ch, temp[64+1]; col = tr->mColumn; if(TrSkipWhitespace(tr)) { col = tr->mColumn; len = 0; ch = tr->mRing[tr->mOut&TR_RING_MASK]; if(ch == '+' || ch == '-') { temp[len] = ch; len++; tr->mOut++; } digis = 0; while(TrLoad(tr)) { ch = tr->mRing[tr->mOut&TR_RING_MASK]; if(!isdigit(ch)) break; if(len < 64) temp[len] = ch; len++; digis++; tr->mOut++; } if(ch == '.') { if(len < 64) temp[len] = ch; len++; tr->mOut++; } while(TrLoad(tr)) { ch = tr->mRing[tr->mOut&TR_RING_MASK]; if(!isdigit(ch)) break; if(len < 64) temp[len] = ch; len++; digis++; tr->mOut++; } if(digis > 0) { if(ch == 'E' || ch == 'e') { if(len < 64) temp[len] = ch; len++; digis = 0; tr->mOut++; if(ch == '+' || ch == '-') { if(len < 64) temp[len] = ch; len++; tr->mOut++; } while(TrLoad(tr)) { ch = tr->mRing[tr->mOut&TR_RING_MASK]; if(!isdigit(ch)) break; if(len < 64) temp[len] = ch; len++; digis++; tr->mOut++; } } tr->mColumn += len; if(digis > 0 && ch != '.' && !isalpha(ch)) { if(len > 64) { TrErrorAt(tr, tr->mLine, col, "Float is too long."); return 0; } temp[len] = '\0'; *value = strtod(temp, nullptr); if(*value < loBound || *value > hiBound) { TrErrorAt(tr, tr->mLine, col, "Expected a value from %f to %f.\n", loBound, hiBound); return 0; } return 1; } } else tr->mColumn += len; } TrErrorAt(tr, tr->mLine, col, "Expected a float.\n"); return 0; } // Reads and validates a string token. static int TrReadString(TokenReaderT *tr, const uint maxLen, char *text) { uint col, len; char ch; col = tr->mColumn; if(TrSkipWhitespace(tr)) { col = tr->mColumn; ch = tr->mRing[tr->mOut&TR_RING_MASK]; if(ch == '\"') { tr->mOut++; len = 0; while(TrLoad(tr)) { ch = tr->mRing[tr->mOut&TR_RING_MASK]; tr->mOut++; if(ch == '\"') break; if(ch == '\n') { TrErrorAt(tr, tr->mLine, col, "Unterminated string at end of line.\n"); return 0; } if(len < maxLen) text[len] = ch; len++; } if(ch != '\"') { tr->mColumn += 1 + len; TrErrorAt(tr, tr->mLine, col, "Unterminated string at end of input.\n"); return 0; } tr->mColumn += 2 + len; if(len > maxLen) { TrErrorAt(tr, tr->mLine, col, "String is too long.\n"); return 0; } text[len] = '\0'; return 1; } } TrErrorAt(tr, tr->mLine, col, "Expected a string.\n"); return 0; } // Reads and validates the given operator. static int TrReadOperator(TokenReaderT *tr, const char *op) { uint col, len; char ch; col = tr->mColumn; if(TrSkipWhitespace(tr)) { col = tr->mColumn; len = 0; while(op[len] != '\0' && TrLoad(tr)) { ch = tr->mRing[tr->mOut&TR_RING_MASK]; if(ch != op[len]) break; len++; tr->mOut++; } tr->mColumn += len; if(op[len] == '\0') return 1; } TrErrorAt(tr, tr->mLine, col, "Expected '%s' operator.\n", op); return 0; } /* Performs a string substitution. Any case-insensitive occurrences of the * pattern string are replaced with the replacement string. The result is * truncated if necessary. */ static int StrSubst(const char *in, const char *pat, const char *rep, const size_t maxLen, char *out) { size_t inLen, patLen, repLen; size_t si, di; int truncated; inLen = strlen(in); patLen = strlen(pat); repLen = strlen(rep); si = 0; di = 0; truncated = 0; while(si < inLen && di < maxLen) { if(patLen <= inLen-si) { if(strncasecmp(&in[si], pat, patLen) == 0) { if(repLen > maxLen-di) { repLen = maxLen - di; truncated = 1; } strncpy(&out[di], rep, repLen); si += patLen; di += repLen; } } out[di] = in[si]; si++; di++; } if(si < inLen) truncated = 1; out[di] = '\0'; return !truncated; } /********************* *** Math routines *** *********************/ // Simple clamp routine. static double Clamp(const double val, const double lower, const double upper) { return std::min(std::max(val, lower), upper); } // Performs linear interpolation. static double Lerp(const double a, const double b, const double f) { return a + f * (b - a); } static inline uint dither_rng(uint *seed) { *seed = *seed * 96314165 + 907633515; return *seed; } // Performs a triangular probability density function dither. The input samples // should be normalized (-1 to +1). static void TpdfDither(double *RESTRICT out, const double *RESTRICT in, const double scale, const int count, const int step, uint *seed) { static constexpr double PRNG_SCALE = 1.0 / std::numeric_limits::max(); for(int i{0};i < count;i++) { uint prn0{dither_rng(seed)}; uint prn1{dither_rng(seed)}; out[i*step] = std::round(in[i]*scale + (prn0*PRNG_SCALE - prn1*PRNG_SCALE)); } } /* Fast Fourier transform routines. The number of points must be a power of * two. */ // Performs bit-reversal ordering. static void FftArrange(const uint n, complex_d *inout) { // Handle in-place arrangement. uint rk{0u}; for(uint k{0u};k < n;k++) { if(rk > k) std::swap(inout[rk], inout[k]); uint m{n}; while(rk&(m >>= 1)) rk &= ~m; rk |= m; } } // Performs the summation. static void FftSummation(const int n, const double s, complex_d *cplx) { double pi; int m, m2; int i, k, mk; pi = s * M_PI; for(m = 1, m2 = 2;m < n; m <<= 1, m2 <<= 1) { // v = Complex (-2.0 * sin (0.5 * pi / m) * sin (0.5 * pi / m), -sin (pi / m)) double sm = sin(0.5 * pi / m); auto v = complex_d{-2.0*sm*sm, -sin(pi / m)}; auto w = complex_d{1.0, 0.0}; for(i = 0;i < m;i++) { for(k = i;k < n;k += m2) { mk = k + m; auto t = w * cplx[mk]; cplx[mk] = cplx[k] - t; cplx[k] = cplx[k] + t; } w += v*w; } } } // Performs a forward FFT. static void FftForward(const uint n, complex_d *inout) { FftArrange(n, inout); FftSummation(n, 1.0, inout); } // Performs an inverse FFT. static void FftInverse(const uint n, complex_d *inout) { FftArrange(n, inout); FftSummation(n, -1.0, inout); double f{1.0 / n}; for(uint i{0};i < n;i++) inout[i] *= f; } /* Calculate the complex helical sequence (or discrete-time analytical signal) * of the given input using the Hilbert transform. Given the natural logarithm * of a signal's magnitude response, the imaginary components can be used as * the angles for minimum-phase reconstruction. */ static void Hilbert(const uint n, complex_d *inout) { uint i; // Handle in-place operation. for(i = 0;i < n;i++) inout[i].imag(0.0); FftInverse(n, inout); for(i = 1;i < (n+1)/2;i++) inout[i] *= 2.0; /* Increment i if n is even. */ i += (n&1)^1; for(;i < n;i++) inout[i] = complex_d{0.0, 0.0}; FftForward(n, inout); } /* Calculate the magnitude response of the given input. This is used in * place of phase decomposition, since the phase residuals are discarded for * minimum phase reconstruction. The mirrored half of the response is also * discarded. */ static void MagnitudeResponse(const uint n, const complex_d *in, double *out) { const uint m = 1 + (n / 2); uint i; for(i = 0;i < m;i++) out[i] = std::max(std::abs(in[i]), EPSILON); } /* Apply a range limit (in dB) to the given magnitude response. This is used * to adjust the effects of the diffuse-field average on the equalization * process. */ static void LimitMagnitudeResponse(const uint n, const uint m, const double limit, const double *in, double *out) { double halfLim; uint i, lower, upper; double ave; halfLim = limit / 2.0; // Convert the response to dB. for(i = 0;i < m;i++) out[i] = 20.0 * std::log10(in[i]); // Use six octaves to calculate the average magnitude of the signal. lower = (static_cast(std::ceil(n / std::pow(2.0, 8.0)))) - 1; upper = (static_cast(std::floor(n / std::pow(2.0, 2.0)))) - 1; ave = 0.0; for(i = lower;i <= upper;i++) ave += out[i]; ave /= upper - lower + 1; // Keep the response within range of the average magnitude. for(i = 0;i < m;i++) out[i] = Clamp(out[i], ave - halfLim, ave + halfLim); // Convert the response back to linear magnitude. for(i = 0;i < m;i++) out[i] = std::pow(10.0, out[i] / 20.0); } /* Reconstructs the minimum-phase component for the given magnitude response * of a signal. This is equivalent to phase recomposition, sans the missing * residuals (which were discarded). The mirrored half of the response is * reconstructed. */ static void MinimumPhase(const uint n, const double *in, complex_d *out) { const uint m = 1 + (n / 2); std::vector mags(n); uint i; for(i = 0;i < m;i++) { mags[i] = std::max(EPSILON, in[i]); out[i] = complex_d{std::log(mags[i]), 0.0}; } for(;i < n;i++) { mags[i] = mags[n - i]; out[i] = out[n - i]; } Hilbert(n, out); // Remove any DC offset the filter has. mags[0] = EPSILON; for(i = 0;i < n;i++) { auto a = std::exp(complex_d{0.0, out[i].imag()}); out[i] = complex_d{mags[i], 0.0} * a; } } /*************************** *** Resampler functions *** ***************************/ /* This is the normalized cardinal sine (sinc) function. * * sinc(x) = { 1, x = 0 * { sin(pi x) / (pi x), otherwise. */ static double Sinc(const double x) { if(std::abs(x) < EPSILON) return 1.0; return std::sin(M_PI * x) / (M_PI * x); } /* The zero-order modified Bessel function of the first kind, used for the * Kaiser window. * * I_0(x) = sum_{k=0}^inf (1 / k!)^2 (x / 2)^(2 k) * = sum_{k=0}^inf ((x / 2)^k / k!)^2 */ static double BesselI_0(const double x) { double term, sum, x2, y, last_sum; int k; // Start at k=1 since k=0 is trivial. term = 1.0; sum = 1.0; x2 = x/2.0; k = 1; // Let the integration converge until the term of the sum is no longer // significant. do { y = x2 / k; k++; last_sum = sum; term *= y * y; sum += term; } while(sum != last_sum); return sum; } /* Calculate a Kaiser window from the given beta value and a normalized k * [-1, 1]. * * w(k) = { I_0(B sqrt(1 - k^2)) / I_0(B), -1 <= k <= 1 * { 0, elsewhere. * * Where k can be calculated as: * * k = i / l, where -l <= i <= l. * * or: * * k = 2 i / M - 1, where 0 <= i <= M. */ static double Kaiser(const double b, const double k) { if(!(k >= -1.0 && k <= 1.0)) return 0.0; return BesselI_0(b * std::sqrt(1.0 - k*k)) / BesselI_0(b); } // Calculates the greatest common divisor of a and b. static uint Gcd(uint x, uint y) { while(y > 0) { uint z{y}; y = x % y; x = z; } return x; } /* Calculates the size (order) of the Kaiser window. Rejection is in dB and * the transition width is normalized frequency (0.5 is nyquist). * * M = { ceil((r - 7.95) / (2.285 2 pi f_t)), r > 21 * { ceil(5.79 / 2 pi f_t), r <= 21. * */ static uint CalcKaiserOrder(const double rejection, const double transition) { double w_t = 2.0 * M_PI * transition; if(rejection > 21.0) return static_cast(std::ceil((rejection - 7.95) / (2.285 * w_t))); return static_cast(std::ceil(5.79 / w_t)); } // Calculates the beta value of the Kaiser window. Rejection is in dB. static double CalcKaiserBeta(const double rejection) { if(rejection > 50.0) return 0.1102 * (rejection - 8.7); if(rejection >= 21.0) return (0.5842 * std::pow(rejection - 21.0, 0.4)) + (0.07886 * (rejection - 21.0)); return 0.0; } /* Calculates a point on the Kaiser-windowed sinc filter for the given half- * width, beta, gain, and cutoff. The point is specified in non-normalized * samples, from 0 to M, where M = (2 l + 1). * * w(k) 2 p f_t sinc(2 f_t x) * * x -- centered sample index (i - l) * k -- normalized and centered window index (x / l) * w(k) -- window function (Kaiser) * p -- gain compensation factor when sampling * f_t -- normalized center frequency (or cutoff; 0.5 is nyquist) */ static double SincFilter(const int l, const double b, const double gain, const double cutoff, const int i) { return Kaiser(b, static_cast(i - l) / l) * 2.0 * gain * cutoff * Sinc(2.0 * cutoff * (i - l)); } /* This is a polyphase sinc-filtered resampler. * * Upsample Downsample * * p/q = 3/2 p/q = 3/5 * * M-+-+-+-> M-+-+-+-> * -------------------+ ---------------------+ * p s * f f f f|f| | p s * f f f f f | * | 0 * 0 0 0|0|0 | | 0 * 0 0 0 0|0| | * v 0 * 0 0|0|0 0 | v 0 * 0 0 0|0|0 | * s * f|f|f f f | s * f f|f|f f | * 0 * |0|0 0 0 0 | 0 * 0|0|0 0 0 | * --------+=+--------+ 0 * |0|0 0 0 0 | * d . d .|d|. d . d ----------+=+--------+ * d . . . .|d|. . . . * q-> * q-+-+-+-> * * P_f(i,j) = q i mod p + pj * P_s(i,j) = floor(q i / p) - j * d[i=0..N-1] = sum_{j=0}^{floor((M - 1) / p)} { * { f[P_f(i,j)] s[P_s(i,j)], P_f(i,j) < M * { 0, P_f(i,j) >= M. } */ // Calculate the resampling metrics and build the Kaiser-windowed sinc filter // that's used to cut frequencies above the destination nyquist. static void ResamplerSetup(ResamplerT *rs, const uint srcRate, const uint dstRate) { double cutoff, width, beta; uint gcd, l; int i; gcd = Gcd(srcRate, dstRate); rs->mP = dstRate / gcd; rs->mQ = srcRate / gcd; /* The cutoff is adjusted by half the transition width, so the transition * ends before the nyquist (0.5). Both are scaled by the downsampling * factor. */ if(rs->mP > rs->mQ) { cutoff = 0.475 / rs->mP; width = 0.05 / rs->mP; } else { cutoff = 0.475 / rs->mQ; width = 0.05 / rs->mQ; } // A rejection of -180 dB is used for the stop band. Round up when // calculating the left offset to avoid increasing the transition width. l = (CalcKaiserOrder(180.0, width)+1) / 2; beta = CalcKaiserBeta(180.0); rs->mM = l*2 + 1; rs->mL = l; rs->mF.resize(rs->mM); for(i = 0;i < (static_cast(rs->mM));i++) rs->mF[i] = SincFilter(static_cast(l), beta, rs->mP, cutoff, i); } // Perform the upsample-filter-downsample resampling operation using a // polyphase filter implementation. static void ResamplerRun(ResamplerT *rs, const uint inN, const double *in, const uint outN, double *out) { const uint p = rs->mP, q = rs->mQ, m = rs->mM, l = rs->mL; std::vector workspace; const double *f = rs->mF.data(); uint j_f, j_s; double *work; uint i; if(outN == 0) return; // Handle in-place operation. if(in == out) { workspace.resize(outN); work = workspace.data(); } else work = out; // Resample the input. for(i = 0;i < outN;i++) { double r = 0.0; // Input starts at l to compensate for the filter delay. This will // drop any build-up from the first half of the filter. j_f = (l + (q * i)) % p; j_s = (l + (q * i)) / p; while(j_f < m) { // Only take input when 0 <= j_s < inN. This single unsigned // comparison catches both cases. if(j_s < inN) r += f[j_f] * in[j_s]; j_f += p; j_s--; } work[i] = r; } // Clean up after in-place operation. if(work != out) { for(i = 0;i < outN;i++) out[i] = work[i]; } } /************************* *** File source input *** *************************/ // Read a binary value of the specified byte order and byte size from a file, // storing it as a 32-bit unsigned integer. static int ReadBin4(FILE *fp, const char *filename, const ByteOrderT order, const uint bytes, uint32_t *out) { uint8_t in[4]; uint32_t accum; uint i; if(fread(in, 1, bytes, fp) != bytes) { fprintf(stderr, "\nError: Bad read from file '%s'.\n", filename); return 0; } accum = 0; switch(order) { case BO_LITTLE: for(i = 0;i < bytes;i++) accum = (accum<<8) | in[bytes - i - 1]; break; case BO_BIG: for(i = 0;i < bytes;i++) accum = (accum<<8) | in[i]; break; default: break; } *out = accum; return 1; } // Read a binary value of the specified byte order from a file, storing it as // a 64-bit unsigned integer. static int ReadBin8(FILE *fp, const char *filename, const ByteOrderT order, uint64_t *out) { uint8_t in[8]; uint64_t accum; uint i; if(fread(in, 1, 8, fp) != 8) { fprintf(stderr, "\nError: Bad read from file '%s'.\n", filename); return 0; } accum = 0ULL; switch(order) { case BO_LITTLE: for(i = 0;i < 8;i++) accum = (accum<<8) | in[8 - i - 1]; break; case BO_BIG: for(i = 0;i < 8;i++) accum = (accum<<8) | in[i]; break; default: break; } *out = accum; return 1; } /* Read a binary value of the specified type, byte order, and byte size from * a file, converting it to a double. For integer types, the significant * bits are used to normalize the result. The sign of bits determines * whether they are padded toward the MSB (negative) or LSB (positive). * Floating-point types are not normalized. */ static int ReadBinAsDouble(FILE *fp, const char *filename, const ByteOrderT order, const ElementTypeT type, const uint bytes, const int bits, double *out) { union { uint32_t ui; int32_t i; float f; } v4; union { uint64_t ui; double f; } v8; *out = 0.0; if(bytes > 4) { if(!ReadBin8(fp, filename, order, &v8.ui)) return 0; if(type == ET_FP) *out = v8.f; } else { if(!ReadBin4(fp, filename, order, bytes, &v4.ui)) return 0; if(type == ET_FP) *out = v4.f; else { if(bits > 0) v4.ui >>= (8*bytes) - (static_cast(bits)); else v4.ui &= (0xFFFFFFFF >> (32+bits)); if(v4.ui&static_cast(1<<(std::abs(bits)-1))) v4.ui |= (0xFFFFFFFF << std::abs(bits)); *out = v4.i / static_cast(1<<(std::abs(bits)-1)); } } return 1; } /* Read an ascii value of the specified type from a file, converting it to a * double. For integer types, the significant bits are used to normalize the * result. The sign of the bits should always be positive. This also skips * up to one separator character before the element itself. */ static int ReadAsciiAsDouble(TokenReaderT *tr, const char *filename, const ElementTypeT type, const uint bits, double *out) { if(TrIsOperator(tr, ",")) TrReadOperator(tr, ","); else if(TrIsOperator(tr, ":")) TrReadOperator(tr, ":"); else if(TrIsOperator(tr, ";")) TrReadOperator(tr, ";"); else if(TrIsOperator(tr, "|")) TrReadOperator(tr, "|"); if(type == ET_FP) { if(!TrReadFloat(tr, -std::numeric_limits::infinity(), std::numeric_limits::infinity(), out)) { fprintf(stderr, "\nError: Bad read from file '%s'.\n", filename); return 0; } } else { int v; if(!TrReadInt(tr, -(1<<(bits-1)), (1<<(bits-1))-1, &v)) { fprintf(stderr, "\nError: Bad read from file '%s'.\n", filename); return 0; } *out = v / static_cast((1<<(bits-1))-1); } return 1; } // Read the RIFF/RIFX WAVE format chunk from a file, validating it against // the source parameters and data set metrics. static int ReadWaveFormat(FILE *fp, const ByteOrderT order, const uint hrirRate, SourceRefT *src) { uint32_t fourCC, chunkSize; uint32_t format, channels, rate, dummy, block, size, bits; chunkSize = 0; do { if(chunkSize > 0) fseek(fp, static_cast(chunkSize), SEEK_CUR); if(!ReadBin4(fp, src->mPath, BO_LITTLE, 4, &fourCC) || !ReadBin4(fp, src->mPath, order, 4, &chunkSize)) return 0; } while(fourCC != FOURCC_FMT); if(!ReadBin4(fp, src->mPath, order, 2, &format) || !ReadBin4(fp, src->mPath, order, 2, &channels) || !ReadBin4(fp, src->mPath, order, 4, &rate) || !ReadBin4(fp, src->mPath, order, 4, &dummy) || !ReadBin4(fp, src->mPath, order, 2, &block)) return 0; block /= channels; if(chunkSize > 14) { if(!ReadBin4(fp, src->mPath, order, 2, &size)) return 0; size /= 8; if(block > size) size = block; } else size = block; if(format == WAVE_FORMAT_EXTENSIBLE) { fseek(fp, 2, SEEK_CUR); if(!ReadBin4(fp, src->mPath, order, 2, &bits)) return 0; if(bits == 0) bits = 8 * size; fseek(fp, 4, SEEK_CUR); if(!ReadBin4(fp, src->mPath, order, 2, &format)) return 0; fseek(fp, static_cast(chunkSize - 26), SEEK_CUR); } else { bits = 8 * size; if(chunkSize > 14) fseek(fp, static_cast(chunkSize - 16), SEEK_CUR); else fseek(fp, static_cast(chunkSize - 14), SEEK_CUR); } if(format != WAVE_FORMAT_PCM && format != WAVE_FORMAT_IEEE_FLOAT) { fprintf(stderr, "\nError: Unsupported WAVE format in file '%s'.\n", src->mPath); return 0; } if(src->mChannel >= channels) { fprintf(stderr, "\nError: Missing source channel in WAVE file '%s'.\n", src->mPath); return 0; } if(rate != hrirRate) { fprintf(stderr, "\nError: Mismatched source sample rate in WAVE file '%s'.\n", src->mPath); return 0; } if(format == WAVE_FORMAT_PCM) { if(size < 2 || size > 4) { fprintf(stderr, "\nError: Unsupported sample size in WAVE file '%s'.\n", src->mPath); return 0; } if(bits < 16 || bits > (8*size)) { fprintf(stderr, "\nError: Bad significant bits in WAVE file '%s'.\n", src->mPath); return 0; } src->mType = ET_INT; } else { if(size != 4 && size != 8) { fprintf(stderr, "\nError: Unsupported sample size in WAVE file '%s'.\n", src->mPath); return 0; } src->mType = ET_FP; } src->mSize = size; src->mBits = static_cast(bits); src->mSkip = channels; return 1; } // Read a RIFF/RIFX WAVE data chunk, converting all elements to doubles. static int ReadWaveData(FILE *fp, const SourceRefT *src, const ByteOrderT order, const uint n, double *hrir) { int pre, post, skip; uint i; pre = static_cast(src->mSize * src->mChannel); post = static_cast(src->mSize * (src->mSkip - src->mChannel - 1)); skip = 0; for(i = 0;i < n;i++) { skip += pre; if(skip > 0) fseek(fp, skip, SEEK_CUR); if(!ReadBinAsDouble(fp, src->mPath, order, src->mType, src->mSize, src->mBits, &hrir[i])) return 0; skip = post; } if(skip > 0) fseek(fp, skip, SEEK_CUR); return 1; } // Read the RIFF/RIFX WAVE list or data chunk, converting all elements to // doubles. static int ReadWaveList(FILE *fp, const SourceRefT *src, const ByteOrderT order, const uint n, double *hrir) { uint32_t fourCC, chunkSize, listSize, count; uint block, skip, offset, i; double lastSample; for(;;) { if(!ReadBin4(fp, src->mPath, BO_LITTLE, 4, &fourCC) || !ReadBin4(fp, src->mPath, order, 4, &chunkSize)) return 0; if(fourCC == FOURCC_DATA) { block = src->mSize * src->mSkip; count = chunkSize / block; if(count < (src->mOffset + n)) { fprintf(stderr, "\nError: Bad read from file '%s'.\n", src->mPath); return 0; } fseek(fp, static_cast(src->mOffset * block), SEEK_CUR); if(!ReadWaveData(fp, src, order, n, &hrir[0])) return 0; return 1; } else if(fourCC == FOURCC_LIST) { if(!ReadBin4(fp, src->mPath, BO_LITTLE, 4, &fourCC)) return 0; chunkSize -= 4; if(fourCC == FOURCC_WAVL) break; } if(chunkSize > 0) fseek(fp, static_cast(chunkSize), SEEK_CUR); } listSize = chunkSize; block = src->mSize * src->mSkip; skip = src->mOffset; offset = 0; lastSample = 0.0; while(offset < n && listSize > 8) { if(!ReadBin4(fp, src->mPath, BO_LITTLE, 4, &fourCC) || !ReadBin4(fp, src->mPath, order, 4, &chunkSize)) return 0; listSize -= 8 + chunkSize; if(fourCC == FOURCC_DATA) { count = chunkSize / block; if(count > skip) { fseek(fp, static_cast(skip * block), SEEK_CUR); chunkSize -= skip * block; count -= skip; skip = 0; if(count > (n - offset)) count = n - offset; if(!ReadWaveData(fp, src, order, count, &hrir[offset])) return 0; chunkSize -= count * block; offset += count; lastSample = hrir[offset - 1]; } else { skip -= count; count = 0; } } else if(fourCC == FOURCC_SLNT) { if(!ReadBin4(fp, src->mPath, order, 4, &count)) return 0; chunkSize -= 4; if(count > skip) { count -= skip; skip = 0; if(count > (n - offset)) count = n - offset; for(i = 0; i < count; i ++) hrir[offset + i] = lastSample; offset += count; } else { skip -= count; count = 0; } } if(chunkSize > 0) fseek(fp, static_cast(chunkSize), SEEK_CUR); } if(offset < n) { fprintf(stderr, "\nError: Bad read from file '%s'.\n", src->mPath); return 0; } return 1; } // Load a source HRIR from an ASCII text file containing a list of elements // separated by whitespace or common list operators (',', ';', ':', '|'). static int LoadAsciiSource(FILE *fp, const SourceRefT *src, const uint n, double *hrir) { TokenReaderT tr; uint i, j; double dummy; TrSetup(fp, nullptr, &tr); for(i = 0;i < src->mOffset;i++) { if(!ReadAsciiAsDouble(&tr, src->mPath, src->mType, static_cast(src->mBits), &dummy)) return 0; } for(i = 0;i < n;i++) { if(!ReadAsciiAsDouble(&tr, src->mPath, src->mType, static_cast(src->mBits), &hrir[i])) return 0; for(j = 0;j < src->mSkip;j++) { if(!ReadAsciiAsDouble(&tr, src->mPath, src->mType, static_cast(src->mBits), &dummy)) return 0; } } return 1; } // Load a source HRIR from a binary file. static int LoadBinarySource(FILE *fp, const SourceRefT *src, const ByteOrderT order, const uint n, double *hrir) { uint i; fseek(fp, static_cast(src->mOffset), SEEK_SET); for(i = 0;i < n;i++) { if(!ReadBinAsDouble(fp, src->mPath, order, src->mType, src->mSize, src->mBits, &hrir[i])) return 0; if(src->mSkip > 0) fseek(fp, static_cast(src->mSkip), SEEK_CUR); } return 1; } // Load a source HRIR from a RIFF/RIFX WAVE file. static int LoadWaveSource(FILE *fp, SourceRefT *src, const uint hrirRate, const uint n, double *hrir) { uint32_t fourCC, dummy; ByteOrderT order; if(!ReadBin4(fp, src->mPath, BO_LITTLE, 4, &fourCC) || !ReadBin4(fp, src->mPath, BO_LITTLE, 4, &dummy)) return 0; if(fourCC == FOURCC_RIFF) order = BO_LITTLE; else if(fourCC == FOURCC_RIFX) order = BO_BIG; else { fprintf(stderr, "\nError: No RIFF/RIFX chunk in file '%s'.\n", src->mPath); return 0; } if(!ReadBin4(fp, src->mPath, BO_LITTLE, 4, &fourCC)) return 0; if(fourCC != FOURCC_WAVE) { fprintf(stderr, "\nError: Not a RIFF/RIFX WAVE file '%s'.\n", src->mPath); return 0; } if(!ReadWaveFormat(fp, order, hrirRate, src)) return 0; if(!ReadWaveList(fp, src, order, n, hrir)) return 0; return 1; } // Load a Spatially Oriented Format for Accoustics (SOFA) file. static struct MYSOFA_EASY* LoadSofaFile(SourceRefT *src, const uint hrirRate, const uint n) { struct MYSOFA_EASY *sofa{mysofa_cache_lookup(src->mPath, (float)hrirRate)}; if(sofa) return sofa; sofa = static_cast(calloc(1, sizeof(*sofa))); if(sofa == nullptr) { fprintf(stderr, "\nError: Out of memory.\n"); return nullptr; } sofa->lookup = nullptr; sofa->neighborhood = nullptr; int err; sofa->hrtf = mysofa_load(src->mPath, &err); if(!sofa->hrtf) { mysofa_close(sofa); fprintf(stderr, "\nError: Could not load source file '%s'.\n", src->mPath); return nullptr; } err = mysofa_check(sofa->hrtf); if(err != MYSOFA_OK) /* NOTE: Some valid SOFA files are failing this check. { mysofa_close(sofa); fprintf(stderr, "\nError: Malformed source file '%s'.\n", src->mPath); return nullptr; }*/ fprintf(stderr, "\nWarning: Supposedly malformed source file '%s'.\n", src->mPath); if((src->mOffset + n) > sofa->hrtf->N) { mysofa_close(sofa); fprintf(stderr, "\nError: Not enough samples in SOFA file '%s'.\n", src->mPath); return nullptr; } if(src->mChannel >= sofa->hrtf->R) { mysofa_close(sofa); fprintf(stderr, "\nError: Missing source receiver in SOFA file '%s'.\n", src->mPath); return nullptr; } mysofa_tocartesian(sofa->hrtf); sofa->lookup = mysofa_lookup_init(sofa->hrtf); if(sofa->lookup == nullptr) { mysofa_close(sofa); fprintf(stderr, "\nError: Out of memory.\n"); return nullptr; } return mysofa_cache_store(sofa, src->mPath, (float)hrirRate); } // Copies the HRIR data from a particular SOFA measurement. static void ExtractSofaHrir(const struct MYSOFA_EASY *sofa, const uint index, const uint channel, const uint offset, const uint n, double *hrir) { for(uint i{0u};i < n;i++) hrir[i] = sofa->hrtf->DataIR.values[(index*sofa->hrtf->R + channel)*sofa->hrtf->N + offset + i]; } // Load a source HRIR from a Spatially Oriented Format for Accoustics (SOFA) // file. static int LoadSofaSource(SourceRefT *src, const uint hrirRate, const uint n, double *hrir) { struct MYSOFA_EASY *sofa; float target[3]; int nearest; float *coords; sofa = LoadSofaFile(src, hrirRate, n); if(sofa == nullptr) return 0; /* NOTE: At some point it may be benficial or necessary to consider the various coordinate systems, listener/source orientations, and direciontal vectors defined in the SOFA file. */ target[0] = src->mAzimuth; target[1] = src->mElevation; target[2] = src->mRadius; mysofa_s2c(target); nearest = mysofa_lookup(sofa->lookup, target); if(nearest < 0) { fprintf(stderr, "\nError: Lookup failed in source file '%s'.\n", src->mPath); return 0; } coords = &sofa->hrtf->SourcePosition.values[3 * nearest]; if(std::fabs(coords[0] - target[0]) > 0.001 || std::fabs(coords[1] - target[1]) > 0.001 || std::fabs(coords[2] - target[2]) > 0.001) { fprintf(stderr, "\nError: No impulse response at coordinates (%.3fr, %.1fev, %.1faz) in file '%s'.\n", src->mRadius, src->mElevation, src->mAzimuth, src->mPath); target[0] = coords[0]; target[1] = coords[1]; target[2] = coords[2]; mysofa_c2s(target); fprintf(stderr, " Nearest candidate at (%.3fr, %.1fev, %.1faz).\n", target[2], target[1], target[0]); return 0; } ExtractSofaHrir(sofa, nearest, src->mChannel, src->mOffset, n, hrir); return 1; } // Load a source HRIR from a supported file type. static int LoadSource(SourceRefT *src, const uint hrirRate, const uint n, double *hrir) { FILE *fp{nullptr}; if(src->mFormat != SF_SOFA) { if(src->mFormat == SF_ASCII) fp = fopen(src->mPath, "r"); else fp = fopen(src->mPath, "rb"); if(fp == nullptr) { fprintf(stderr, "\nError: Could not open source file '%s'.\n", src->mPath); return 0; } } int result; switch(src->mFormat) { case SF_ASCII: result = LoadAsciiSource(fp, src, n, hrir); break; case SF_BIN_LE: result = LoadBinarySource(fp, src, BO_LITTLE, n, hrir); break; case SF_BIN_BE: result = LoadBinarySource(fp, src, BO_BIG, n, hrir); break; case SF_WAVE: result = LoadWaveSource(fp, src, hrirRate, n, hrir); break; case SF_SOFA: result = LoadSofaSource(src, hrirRate, n, hrir); break; default: result = 0; } if(fp) fclose(fp); return result; } /*************************** *** File storage output *** ***************************/ // Write an ASCII string to a file. static int WriteAscii(const char *out, FILE *fp, const char *filename) { size_t len; len = strlen(out); if(fwrite(out, 1, len, fp) != len) { fclose(fp); fprintf(stderr, "\nError: Bad write to file '%s'.\n", filename); return 0; } return 1; } // Write a binary value of the given byte order and byte size to a file, // loading it from a 32-bit unsigned integer. static int WriteBin4(const ByteOrderT order, const uint bytes, const uint32_t in, FILE *fp, const char *filename) { uint8_t out[4]; uint i; switch(order) { case BO_LITTLE: for(i = 0;i < bytes;i++) out[i] = (in>>(i*8)) & 0x000000FF; break; case BO_BIG: for(i = 0;i < bytes;i++) out[bytes - i - 1] = (in>>(i*8)) & 0x000000FF; break; default: break; } if(fwrite(out, 1, bytes, fp) != bytes) { fprintf(stderr, "\nError: Bad write to file '%s'.\n", filename); return 0; } return 1; } // Store the OpenAL Soft HRTF data set. static int StoreMhr(const HrirDataT *hData, const char *filename) { uint channels = (hData->mChannelType == CT_STEREO) ? 2 : 1; uint n = hData->mIrPoints; FILE *fp; uint fi, ei, ai, i; uint dither_seed = 22222; if((fp=fopen(filename, "wb")) == nullptr) { fprintf(stderr, "\nError: Could not open MHR file '%s'.\n", filename); return 0; } if(!WriteAscii(MHR_FORMAT, fp, filename)) return 0; if(!WriteBin4(BO_LITTLE, 4, hData->mIrRate, fp, filename)) return 0; if(!WriteBin4(BO_LITTLE, 1, static_cast(hData->mSampleType), fp, filename)) return 0; if(!WriteBin4(BO_LITTLE, 1, static_cast(hData->mChannelType), fp, filename)) return 0; if(!WriteBin4(BO_LITTLE, 1, hData->mIrPoints, fp, filename)) return 0; if(!WriteBin4(BO_LITTLE, 1, hData->mFdCount, fp, filename)) return 0; for(fi = 0;fi < hData->mFdCount;fi++) { auto fdist = static_cast(std::round(1000.0 * hData->mFds[fi].mDistance)); if(!WriteBin4(BO_LITTLE, 2, fdist, fp, filename)) return 0; if(!WriteBin4(BO_LITTLE, 1, hData->mFds[fi].mEvCount, fp, filename)) return 0; for(ei = 0;ei < hData->mFds[fi].mEvCount;ei++) { if(!WriteBin4(BO_LITTLE, 1, hData->mFds[fi].mEvs[ei].mAzCount, fp, filename)) return 0; } } for(fi = 0;fi < hData->mFdCount;fi++) { const double scale = (hData->mSampleType == ST_S16) ? 32767.0 : ((hData->mSampleType == ST_S24) ? 8388607.0 : 0.0); const int bps = (hData->mSampleType == ST_S16) ? 2 : ((hData->mSampleType == ST_S24) ? 3 : 0); for(ei = 0;ei < hData->mFds[fi].mEvCount;ei++) { for(ai = 0;ai < hData->mFds[fi].mEvs[ei].mAzCount;ai++) { HrirAzT *azd = &hData->mFds[fi].mEvs[ei].mAzs[ai]; double out[2 * MAX_TRUNCSIZE]; TpdfDither(out, azd->mIrs[0], scale, n, channels, &dither_seed); if(hData->mChannelType == CT_STEREO) TpdfDither(out+1, azd->mIrs[1], scale, n, channels, &dither_seed); for(i = 0;i < (channels * n);i++) { int v = static_cast(Clamp(out[i], -scale-1.0, scale)); if(!WriteBin4(BO_LITTLE, bps, static_cast(v), fp, filename)) return 0; } } } } for(fi = 0;fi < hData->mFdCount;fi++) { for(ei = 0;ei < hData->mFds[fi].mEvCount;ei++) { for(ai = 0;ai < hData->mFds[fi].mEvs[ei].mAzCount;ai++) { const HrirAzT &azd = hData->mFds[fi].mEvs[ei].mAzs[ai]; int v = static_cast(std::min(std::round(hData->mIrRate * azd.mDelays[0]), MAX_HRTD)); if(!WriteBin4(BO_LITTLE, 1, static_cast(v), fp, filename)) return 0; if(hData->mChannelType == CT_STEREO) { v = static_cast(std::min(std::round(hData->mIrRate * azd.mDelays[1]), MAX_HRTD)); if(!WriteBin4(BO_LITTLE, 1, static_cast(v), fp, filename)) return 0; } } } } fclose(fp); return 1; } /*********************** *** HRTF processing *** ***********************/ // Calculate the onset time of an HRIR and average it with any existing // timing for its field, elevation, azimuth, and ear. static double AverageHrirOnset(const uint rate, const uint n, const double *hrir, const double f, const double onset) { std::vector upsampled(10 * n); { ResamplerT rs; ResamplerSetup(&rs, rate, 10 * rate); ResamplerRun(&rs, n, hrir, 10 * n, upsampled.data()); } double mag{0.0}; for(uint i{0u};i < 10*n;i++) mag = std::max(std::abs(upsampled[i]), mag); mag *= 0.15; uint i{0u}; for(;i < 10*n;i++) { if(std::abs(upsampled[i]) >= mag) break; } return Lerp(onset, static_cast(i) / (10*rate), f); } // Calculate the magnitude response of an HRIR and average it with any // existing responses for its field, elevation, azimuth, and ear. static void AverageHrirMagnitude(const uint points, const uint n, const double *hrir, const double f, double *mag) { uint m = 1 + (n / 2), i; std::vector h(n); std::vector r(n); for(i = 0;i < points;i++) h[i] = complex_d{hrir[i], 0.0}; for(;i < n;i++) h[i] = complex_d{0.0, 0.0}; FftForward(n, h.data()); MagnitudeResponse(n, h.data(), r.data()); for(i = 0;i < m;i++) mag[i] = Lerp(mag[i], r[i], f); } /* Balances the maximum HRIR magnitudes of multi-field data sets by * independently normalizing each field in relation to the overall maximum. * This is done to ignore distance attenuation. */ static void BalanceFieldMagnitudes(const HrirDataT *hData, const uint channels, const uint m) { double maxMags[MAX_FD_COUNT]; uint fi, ei, ai, ti, i; double maxMag{0.0}; for(fi = 0;fi < hData->mFdCount;fi++) { maxMags[fi] = 0.0; for(ei = hData->mFds[fi].mEvStart;ei < hData->mFds[fi].mEvCount;ei++) { for(ai = 0;ai < hData->mFds[fi].mEvs[ei].mAzCount;ai++) { HrirAzT *azd = &hData->mFds[fi].mEvs[ei].mAzs[ai]; for(ti = 0;ti < channels;ti++) { for(i = 0;i < m;i++) maxMags[fi] = std::max(azd->mIrs[ti][i], maxMags[fi]); } } } maxMag = std::max(maxMags[fi], maxMag); } for(fi = 0;fi < hData->mFdCount;fi++) { maxMags[fi] /= maxMag; for(ei = hData->mFds[fi].mEvStart;ei < hData->mFds[fi].mEvCount;ei++) { for(ai = 0;ai < hData->mFds[fi].mEvs[ei].mAzCount;ai++) { HrirAzT *azd = &hData->mFds[fi].mEvs[ei].mAzs[ai]; for(ti = 0;ti < channels;ti++) { for(i = 0;i < m;i++) azd->mIrs[ti][i] /= maxMags[fi]; } } } } } /* Calculate the contribution of each HRIR to the diffuse-field average based * on its coverage volume. All volumes are centered at the spherical HRIR * coordinates and measured by extruded solid angle. */ static void CalculateDfWeights(const HrirDataT *hData, double *weights) { double sum, innerRa, outerRa, evs, ev, upperEv, lowerEv; double solidAngle, solidVolume; uint fi, ei; sum = 0.0; // The head radius acts as the limit for the inner radius. innerRa = hData->mRadius; for(fi = 0;fi < hData->mFdCount;fi++) { // Each volume ends half way between progressive field measurements. if((fi + 1) < hData->mFdCount) outerRa = 0.5f * (hData->mFds[fi].mDistance + hData->mFds[fi + 1].mDistance); // The final volume has its limit extended to some practical value. // This is done to emphasize the far-field responses in the average. else outerRa = 10.0f; evs = M_PI / 2.0 / (hData->mFds[fi].mEvCount - 1); for(ei = hData->mFds[fi].mEvStart;ei < hData->mFds[fi].mEvCount;ei++) { // For each elevation, calculate the upper and lower limits of // the patch band. ev = hData->mFds[fi].mEvs[ei].mElevation; lowerEv = std::max(-M_PI / 2.0, ev - evs); upperEv = std::min(M_PI / 2.0, ev + evs); // Calculate the surface area of the patch band. solidAngle = 2.0 * M_PI * (std::sin(upperEv) - std::sin(lowerEv)); // Then the volume of the extruded patch band. solidVolume = solidAngle * (std::pow(outerRa, 3.0) - std::pow(innerRa, 3.0)) / 3.0; // Each weight is the volume of one extruded patch. weights[(fi * MAX_EV_COUNT) + ei] = solidVolume / hData->mFds[fi].mEvs[ei].mAzCount; // Sum the total coverage volume of the HRIRs for all fields. sum += solidAngle; } innerRa = outerRa; } for(fi = 0;fi < hData->mFdCount;fi++) { // Normalize the weights given the total surface coverage for all // fields. for(ei = hData->mFds[fi].mEvStart;ei < hData->mFds[fi].mEvCount;ei++) weights[(fi * MAX_EV_COUNT) + ei] /= sum; } } /* Calculate the diffuse-field average from the given magnitude responses of * the HRIR set. Weighting can be applied to compensate for the varying * coverage of each HRIR. The final average can then be limited by the * specified magnitude range (in positive dB; 0.0 to skip). */ static void CalculateDiffuseFieldAverage(const HrirDataT *hData, const uint channels, const uint m, const int weighted, const double limit, double *dfa) { std::vector weights(hData->mFdCount * MAX_EV_COUNT); uint count, ti, fi, ei, i, ai; if(weighted) { // Use coverage weighting to calculate the average. CalculateDfWeights(hData, weights.data()); } else { double weight; // If coverage weighting is not used, the weights still need to be // averaged by the number of existing HRIRs. count = hData->mIrCount; for(fi = 0;fi < hData->mFdCount;fi++) { for(ei = 0;ei < hData->mFds[fi].mEvStart;ei++) count -= hData->mFds[fi].mEvs[ei].mAzCount; } weight = 1.0 / count; for(fi = 0;fi < hData->mFdCount;fi++) { for(ei = hData->mFds[fi].mEvStart;ei < hData->mFds[fi].mEvCount;ei++) weights[(fi * MAX_EV_COUNT) + ei] = weight; } } for(ti = 0;ti < channels;ti++) { for(i = 0;i < m;i++) dfa[(ti * m) + i] = 0.0; for(fi = 0;fi < hData->mFdCount;fi++) { for(ei = hData->mFds[fi].mEvStart;ei < hData->mFds[fi].mEvCount;ei++) { for(ai = 0;ai < hData->mFds[fi].mEvs[ei].mAzCount;ai++) { HrirAzT *azd = &hData->mFds[fi].mEvs[ei].mAzs[ai]; // Get the weight for this HRIR's contribution. double weight = weights[(fi * MAX_EV_COUNT) + ei]; // Add this HRIR's weighted power average to the total. for(i = 0;i < m;i++) dfa[(ti * m) + i] += weight * azd->mIrs[ti][i] * azd->mIrs[ti][i]; } } } // Finish the average calculation and keep it from being too small. for(i = 0;i < m;i++) dfa[(ti * m) + i] = std::max(sqrt(dfa[(ti * m) + i]), EPSILON); // Apply a limit to the magnitude range of the diffuse-field average // if desired. if(limit > 0.0) LimitMagnitudeResponse(hData->mFftSize, m, limit, &dfa[ti * m], &dfa[ti * m]); } } // Perform diffuse-field equalization on the magnitude responses of the HRIR // set using the given average response. static void DiffuseFieldEqualize(const uint channels, const uint m, const double *dfa, const HrirDataT *hData) { uint ti, fi, ei, ai, i; for(fi = 0;fi < hData->mFdCount;fi++) { for(ei = hData->mFds[fi].mEvStart;ei < hData->mFds[fi].mEvCount;ei++) { for(ai = 0;ai < hData->mFds[fi].mEvs[ei].mAzCount;ai++) { HrirAzT *azd = &hData->mFds[fi].mEvs[ei].mAzs[ai]; for(ti = 0;ti < channels;ti++) { for(i = 0;i < m;i++) azd->mIrs[ti][i] /= dfa[(ti * m) + i]; } } } } } /* Perform minimum-phase reconstruction using the magnitude responses of the * HRIR set. Work is delegated to this struct, which runs asynchronously on one * or more threads (sharing the same reconstructor object). */ struct HrirReconstructor { std::vector mIrs; std::atomic mCurrent; std::atomic mDone; size_t mFftSize; size_t mIrPoints; void Worker() { auto h = std::vector(mFftSize); while(1) { /* Load the current index to process. */ size_t idx{mCurrent.load()}; do { /* If the index is at the end, we're done. */ if(idx >= mIrs.size()) return; /* Otherwise, increment the current index atomically so other * threads know to go to the next one. If this call fails, the * current index was just changed by another thread and the new * value is loaded into idx, which we'll recheck. */ } while(!mCurrent.compare_exchange_weak(idx, idx+1, std::memory_order_relaxed)); /* Now do the reconstruction, and apply the inverse FFT to get the * time-domain response. */ MinimumPhase(mFftSize, mIrs[idx], h.data()); FftInverse(mFftSize, h.data()); for(size_t i{0u};i < mIrPoints;++i) mIrs[idx][i] = h[i].real(); /* Increment the number of IRs done. */ mDone.fetch_add(1); } } }; static void ReconstructHrirs(const HrirDataT *hData) { const uint channels{(hData->mChannelType == CT_STEREO) ? 2u : 1u}; /* Count the number of IRs to process (excluding elevations that will be * synthesized later). */ size_t total{hData->mIrCount}; for(uint fi{0u};fi < hData->mFdCount;fi++) { for(uint ei{0u};ei < hData->mFds[fi].mEvStart;ei++) total -= hData->mFds[fi].mEvs[ei].mAzCount; } total *= channels; /* Set up the reconstructor with the needed size info and pointers to the * IRs to process. */ HrirReconstructor reconstructor; reconstructor.mIrs.reserve(total); reconstructor.mCurrent.store(0, std::memory_order_relaxed); reconstructor.mDone.store(0, std::memory_order_relaxed); reconstructor.mFftSize = hData->mFftSize; reconstructor.mIrPoints = hData->mIrPoints; for(uint fi{0u};fi < hData->mFdCount;fi++) { const HrirFdT &field = hData->mFds[fi]; for(uint ei{field.mEvStart};ei < field.mEvCount;ei++) { const HrirEvT &elev = field.mEvs[ei]; for(uint ai{0u};ai < elev.mAzCount;ai++) { const HrirAzT &azd = elev.mAzs[ai]; for(uint ti{0u};ti < channels;ti++) reconstructor.mIrs.push_back(azd.mIrs[ti]); } } } /* Launch two threads to work on reconstruction. */ std::thread thrd1{std::mem_fn(&HrirReconstructor::Worker), &reconstructor}; std::thread thrd2{std::mem_fn(&HrirReconstructor::Worker), &reconstructor}; /* Keep track of the number of IRs done, periodically reporting it. */ size_t count; while((count=reconstructor.mDone.load()) != total) { size_t pcdone{count * 100 / total}; printf("\r%3zu%% done (%zu of %zu)", pcdone, count, total); fflush(stdout); std::this_thread::sleep_for(std::chrono::milliseconds{50}); } size_t pcdone{count * 100 / total}; printf("\r%3zu%% done (%zu of %zu)\n", pcdone, count, total); if(thrd2.joinable()) thrd2.join(); if(thrd1.joinable()) thrd1.join(); } // Resamples the HRIRs for use at the given sampling rate. static void ResampleHrirs(const uint rate, HrirDataT *hData) { uint channels = (hData->mChannelType == CT_STEREO) ? 2 : 1; uint n = hData->mIrPoints; uint ti, fi, ei, ai; ResamplerT rs; ResamplerSetup(&rs, hData->mIrRate, rate); for(fi = 0;fi < hData->mFdCount;fi++) { for(ei = hData->mFds[fi].mEvStart;ei < hData->mFds[fi].mEvCount;ei++) { for(ai = 0;ai < hData->mFds[fi].mEvs[ei].mAzCount;ai++) { HrirAzT *azd = &hData->mFds[fi].mEvs[ei].mAzs[ai]; for(ti = 0;ti < channels;ti++) ResamplerRun(&rs, n, azd->mIrs[ti], n, azd->mIrs[ti]); } } } hData->mIrRate = rate; } /* Given field and elevation indices and an azimuth, calculate the indices of * the two HRIRs that bound the coordinate along with a factor for * calculating the continuous HRIR using interpolation. */ static void CalcAzIndices(const HrirFdT &field, const uint ei, const double az, uint *a0, uint *a1, double *af) { double f{(2.0*M_PI + az) * field.mEvs[ei].mAzCount / (2.0*M_PI)}; uint i{static_cast(f) % field.mEvs[ei].mAzCount}; f -= std::floor(f); *a0 = i; *a1 = (i + 1) % field.mEvs[ei].mAzCount; *af = f; } /* Synthesize any missing onset timings at the bottom elevations of each * field. This just mirrors the top elevation for the bottom, and blends the * remaining elevations (not an accurate model). */ static void SynthesizeOnsets(HrirDataT *hData) { const uint channels{(hData->mChannelType == CT_STEREO) ? 2u : 1u}; auto proc_field = [channels](HrirFdT &field) -> void { const uint oi{field.mEvStart}; if(oi <= 0) return; /* Get the -90 degree elevation delays by mirroring the +90 degree * elevation delays. The responses are on a spherical grid centered * between the ears, so these should align. */ if(channels > 1) { field.mEvs[0].mAzs[0].mDelays[0] = field.mEvs[field.mEvCount-1].mAzs[0].mDelays[1]; field.mEvs[0].mAzs[0].mDelays[1] = field.mEvs[field.mEvCount-1].mAzs[0].mDelays[0]; } else field.mEvs[0].mAzs[0].mDelays[0] = field.mEvs[field.mEvCount-1].mAzs[0].mDelays[0]; for(uint ei{1u};ei < field.mEvStart;ei++) { const double of{static_cast(ei) / field.mEvStart}; for(uint ai{0u};ai < field.mEvs[ei].mAzCount;ai++) { uint a0, a1; double af; CalcAzIndices(field, oi, field.mEvs[ei].mAzs[ai].mAzimuth, &a0, &a1, &af); for(uint ti{0u};ti < channels;ti++) { double other{Lerp(field.mEvs[oi].mAzs[a0].mDelays[ti], field.mEvs[oi].mAzs[a1].mDelays[ti], af)}; field.mEvs[ei].mAzs[ai].mDelays[ti] = Lerp(field.mEvs[0].mAzs[0].mDelays[ti], other, of); } } } }; std::for_each(hData->mFds.begin(), hData->mFds.begin()+hData->mFdCount, proc_field); } /* Attempt to synthesize any missing HRIRs at the bottom elevations of each * field. Right now this just blends the lowest elevation HRIRs together and * applies some attenuation and high frequency damping. It is a simple, if * inaccurate model. */ static void SynthesizeHrirs(HrirDataT *hData) { const uint channels{(hData->mChannelType == CT_STEREO) ? 2u : 1u}; const uint irSize{hData->mIrPoints}; const double beta{3.5e-6 * hData->mIrRate}; auto proc_field = [channels,irSize,beta](HrirFdT &field) -> void { const uint oi{field.mEvStart}; if(oi <= 0) return; for(uint ti{0u};ti < channels;ti++) { for(uint i{0u};i < irSize;i++) field.mEvs[0].mAzs[0].mIrs[ti][i] = 0.0; /* Blend the lowest defined elevation's responses for an average * -90 degree elevation response. */ double blend_count{0.0}; for(uint ai{0u};ai < field.mEvs[oi].mAzCount;ai++) { /* Only include the left responses for the left ear, and the * right responses for the right ear. This removes the cross- * talk that shouldn't exist for the -90 degree elevation * response (and would be mistimed anyway). NOTE: Azimuth goes * from 0...2pi rather than -pi...+pi (0 in front, clockwise). */ if(std::abs(field.mEvs[oi].mAzs[ai].mAzimuth) < EPSILON || (ti == LeftChannel && field.mEvs[oi].mAzs[ai].mAzimuth > M_PI-EPSILON) || (ti == RightChannel && field.mEvs[oi].mAzs[ai].mAzimuth < M_PI+EPSILON)) { for(uint i{0u};i < irSize;i++) field.mEvs[0].mAzs[0].mIrs[ti][i] += field.mEvs[oi].mAzs[ai].mIrs[ti][i]; blend_count += 1.0; } } if(blend_count > 0.0) { for(uint i{0u};i < irSize;i++) field.mEvs[0].mAzs[0].mIrs[ti][i] /= blend_count; } for(uint ei{1u};ei < field.mEvStart;ei++) { const double of{static_cast(ei) / field.mEvStart}; const double b{(1.0 - of) * beta}; for(uint ai{0u};ai < field.mEvs[ei].mAzCount;ai++) { uint a0, a1; double af; CalcAzIndices(field, oi, field.mEvs[ei].mAzs[ai].mAzimuth, &a0, &a1, &af); double lp[4]{}; for(uint i{0u};i < irSize;i++) { /* Blend the two defined HRIRs closest to this azimuth, * then blend that with the synthesized -90 elevation. */ const double s1{Lerp(field.mEvs[oi].mAzs[a0].mIrs[ti][i], field.mEvs[oi].mAzs[a1].mIrs[ti][i], af)}; const double s0{Lerp(field.mEvs[0].mAzs[0].mIrs[ti][i], s1, of)}; /* Apply a low-pass to simulate body occlusion. */ lp[0] = Lerp(s0, lp[0], b); lp[1] = Lerp(lp[0], lp[1], b); lp[2] = Lerp(lp[1], lp[2], b); lp[3] = Lerp(lp[2], lp[3], b); field.mEvs[ei].mAzs[ai].mIrs[ti][i] = lp[3]; } } } const double b{beta}; double lp[4]{}; for(uint i{0u};i < irSize;i++) { const double s0{field.mEvs[0].mAzs[0].mIrs[ti][i]}; lp[0] = Lerp(s0, lp[0], b); lp[1] = Lerp(lp[0], lp[1], b); lp[2] = Lerp(lp[1], lp[2], b); lp[3] = Lerp(lp[2], lp[3], b); field.mEvs[0].mAzs[0].mIrs[ti][i] = lp[3]; } } field.mEvStart = 0; }; std::for_each(hData->mFds.begin(), hData->mFds.begin()+hData->mFdCount, proc_field); } // The following routines assume a full set of HRIRs for all elevations. // Normalize the HRIR set and slightly attenuate the result. static void NormalizeHrirs(HrirDataT *hData) { const uint channels{(hData->mChannelType == CT_STEREO) ? 2u : 1u}; const uint irSize{hData->mIrPoints}; /* Find the maximum amplitude and RMS out of all the IRs. */ struct LevelPair { double amp, rms; }; auto proc0_field = [channels,irSize](const LevelPair levels, const HrirFdT &field) -> LevelPair { auto proc_elev = [channels,irSize](const LevelPair levels, const HrirEvT &elev) -> LevelPair { auto proc_azi = [channels,irSize](const LevelPair levels, const HrirAzT &azi) -> LevelPair { auto proc_channel = [irSize](const LevelPair levels, const double *ir) -> LevelPair { /* Calculate the peak amplitude and RMS of this IR. */ auto current = std::accumulate(ir, ir+irSize, LevelPair{0.0, 0.0}, [](const LevelPair current, const double impulse) -> LevelPair { return LevelPair{std::max(std::abs(impulse), current.amp), current.rms + impulse*impulse}; }); current.rms = std::sqrt(current.rms / irSize); /* Accumulate levels by taking the maximum amplitude and RMS. */ return LevelPair{std::max(current.amp, levels.amp), std::max(current.rms, levels.rms)}; }; return std::accumulate(azi.mIrs, azi.mIrs+channels, levels, proc_channel); }; return std::accumulate(elev.mAzs, elev.mAzs+elev.mAzCount, levels, proc_azi); }; return std::accumulate(field.mEvs, field.mEvs+field.mEvCount, levels, proc_elev); }; const auto maxlev = std::accumulate(hData->mFds.begin(), hData->mFds.begin()+hData->mFdCount, LevelPair{0.0, 0.0}, proc0_field); /* Normalize using the maximum RMS of the HRIRs. The RMS measure for the * non-filtered signal is of an impulse with equal length (to the filter): * * rms_impulse = sqrt(sum([ 1^2, 0^2, 0^2, ... ]) / n) * = sqrt(1 / n) * * This helps keep a more consistent volume between the non-filtered signal * and various data sets. */ double factor{std::sqrt(1.0 / irSize) / maxlev.rms}; /* Also ensure the samples themselves won't clip. */ factor = std::min(factor, 0.99/maxlev.amp); /* Now scale all IRs by the given factor. */ auto proc1_field = [channels,irSize,factor](HrirFdT &field) -> void { auto proc_elev = [channels,irSize,factor](HrirEvT &elev) -> void { auto proc_azi = [channels,irSize,factor](HrirAzT &azi) -> void { auto proc_channel = [irSize,factor](double *ir) -> void { std::transform(ir, ir+irSize, ir, std::bind(std::multiplies{}, _1, factor)); }; std::for_each(azi.mIrs, azi.mIrs+channels, proc_channel); }; std::for_each(elev.mAzs, elev.mAzs+elev.mAzCount, proc_azi); }; std::for_each(field.mEvs, field.mEvs+field.mEvCount, proc_elev); }; std::for_each(hData->mFds.begin(), hData->mFds.begin()+hData->mFdCount, proc1_field); } // Calculate the left-ear time delay using a spherical head model. static double CalcLTD(const double ev, const double az, const double rad, const double dist) { double azp, dlp, l, al; azp = std::asin(std::cos(ev) * std::sin(az)); dlp = std::sqrt((dist*dist) + (rad*rad) + (2.0*dist*rad*sin(azp))); l = std::sqrt((dist*dist) - (rad*rad)); al = (0.5 * M_PI) + azp; if(dlp > l) dlp = l + (rad * (al - std::acos(rad / dist))); return dlp / 343.3; } // Calculate the effective head-related time delays for each minimum-phase // HRIR. This is done per-field since distance delay is ignored. static void CalculateHrtds(const HeadModelT model, const double radius, HrirDataT *hData) { uint channels = (hData->mChannelType == CT_STEREO) ? 2 : 1; double customRatio{radius / hData->mRadius}; uint ti, fi, ei, ai; if(model == HM_SPHERE) { for(fi = 0;fi < hData->mFdCount;fi++) { for(ei = 0;ei < hData->mFds[fi].mEvCount;ei++) { HrirEvT *evd = &hData->mFds[fi].mEvs[ei]; for(ai = 0;ai < evd->mAzCount;ai++) { HrirAzT *azd = &evd->mAzs[ai]; for(ti = 0;ti < channels;ti++) azd->mDelays[ti] = CalcLTD(evd->mElevation, azd->mAzimuth, radius, hData->mFds[fi].mDistance); } } } } else if(customRatio != 1.0) { for(fi = 0;fi < hData->mFdCount;fi++) { for(ei = 0;ei < hData->mFds[fi].mEvCount;ei++) { HrirEvT *evd = &hData->mFds[fi].mEvs[ei]; for(ai = 0;ai < evd->mAzCount;ai++) { HrirAzT *azd = &evd->mAzs[ai]; for(ti = 0;ti < channels;ti++) azd->mDelays[ti] *= customRatio; } } } } for(fi = 0;fi < hData->mFdCount;fi++) { double minHrtd{std::numeric_limits::infinity()}; for(ei = 0;ei < hData->mFds[fi].mEvCount;ei++) { for(ai = 0;ai < hData->mFds[fi].mEvs[ei].mAzCount;ai++) { HrirAzT *azd = &hData->mFds[fi].mEvs[ei].mAzs[ai]; for(ti = 0;ti < channels;ti++) minHrtd = std::min(azd->mDelays[ti], minHrtd); } } for(ei = 0;ei < hData->mFds[fi].mEvCount;ei++) { for(ti = 0;ti < channels;ti++) { for(ai = 0;ai < hData->mFds[fi].mEvs[ei].mAzCount;ai++) hData->mFds[fi].mEvs[ei].mAzs[ai].mDelays[ti] -= minHrtd; } } } } // Allocate and configure dynamic HRIR structures. static int PrepareHrirData(const uint fdCount, const double distances[MAX_FD_COUNT], const uint evCounts[MAX_FD_COUNT], const uint azCounts[MAX_FD_COUNT * MAX_EV_COUNT], HrirDataT *hData) { uint evTotal = 0, azTotal = 0, fi, ei, ai; for(fi = 0;fi < fdCount;fi++) { evTotal += evCounts[fi]; for(ei = 0;ei < evCounts[fi];ei++) azTotal += azCounts[(fi * MAX_EV_COUNT) + ei]; } if(!fdCount || !evTotal || !azTotal) return 0; hData->mEvsBase.resize(evTotal); hData->mAzsBase.resize(azTotal); hData->mFds.resize(fdCount); hData->mIrCount = azTotal; hData->mFdCount = fdCount; evTotal = 0; azTotal = 0; for(fi = 0;fi < fdCount;fi++) { hData->mFds[fi].mDistance = distances[fi]; hData->mFds[fi].mEvCount = evCounts[fi]; hData->mFds[fi].mEvStart = 0; hData->mFds[fi].mEvs = &hData->mEvsBase[evTotal]; evTotal += evCounts[fi]; for(ei = 0;ei < evCounts[fi];ei++) { uint azCount = azCounts[(fi * MAX_EV_COUNT) + ei]; hData->mFds[fi].mIrCount += azCount; hData->mFds[fi].mEvs[ei].mElevation = -M_PI / 2.0 + M_PI * ei / (evCounts[fi] - 1); hData->mFds[fi].mEvs[ei].mIrCount += azCount; hData->mFds[fi].mEvs[ei].mAzCount = azCount; hData->mFds[fi].mEvs[ei].mAzs = &hData->mAzsBase[azTotal]; for(ai = 0;ai < azCount;ai++) { hData->mFds[fi].mEvs[ei].mAzs[ai].mAzimuth = 2.0 * M_PI * ai / azCount; hData->mFds[fi].mEvs[ei].mAzs[ai].mIndex = azTotal + ai; hData->mFds[fi].mEvs[ei].mAzs[ai].mDelays[0] = 0.0; hData->mFds[fi].mEvs[ei].mAzs[ai].mDelays[1] = 0.0; hData->mFds[fi].mEvs[ei].mAzs[ai].mIrs[0] = nullptr; hData->mFds[fi].mEvs[ei].mAzs[ai].mIrs[1] = nullptr; } azTotal += azCount; } } return 1; } // Match the channel type from a given identifier. static ChannelTypeT MatchChannelType(const char *ident) { if(strcasecmp(ident, "mono") == 0) return CT_MONO; if(strcasecmp(ident, "stereo") == 0) return CT_STEREO; return CT_NONE; } // Process the data set definition to read and validate the data set metrics. static int ProcessMetrics(TokenReaderT *tr, const uint fftSize, const uint truncSize, HrirDataT *hData) { int hasRate = 0, hasType = 0, hasPoints = 0, hasRadius = 0; int hasDistance = 0, hasAzimuths = 0; char ident[MAX_IDENT_LEN+1]; uint line, col; double fpVal; uint points; int intVal; double distances[MAX_FD_COUNT]; uint fdCount = 0; uint evCounts[MAX_FD_COUNT]; std::vector azCounts(MAX_FD_COUNT * MAX_EV_COUNT); TrIndication(tr, &line, &col); while(TrIsIdent(tr)) { TrIndication(tr, &line, &col); if(!TrReadIdent(tr, MAX_IDENT_LEN, ident)) return 0; if(strcasecmp(ident, "rate") == 0) { if(hasRate) { TrErrorAt(tr, line, col, "Redefinition of 'rate'.\n"); return 0; } if(!TrReadOperator(tr, "=")) return 0; if(!TrReadInt(tr, MIN_RATE, MAX_RATE, &intVal)) return 0; hData->mIrRate = static_cast(intVal); hasRate = 1; } else if(strcasecmp(ident, "type") == 0) { char type[MAX_IDENT_LEN+1]; if(hasType) { TrErrorAt(tr, line, col, "Redefinition of 'type'.\n"); return 0; } if(!TrReadOperator(tr, "=")) return 0; if(!TrReadIdent(tr, MAX_IDENT_LEN, type)) return 0; hData->mChannelType = MatchChannelType(type); if(hData->mChannelType == CT_NONE) { TrErrorAt(tr, line, col, "Expected a channel type.\n"); return 0; } hasType = 1; } else if(strcasecmp(ident, "points") == 0) { if(hasPoints) { TrErrorAt(tr, line, col, "Redefinition of 'points'.\n"); return 0; } if(!TrReadOperator(tr, "=")) return 0; TrIndication(tr, &line, &col); if(!TrReadInt(tr, MIN_POINTS, MAX_POINTS, &intVal)) return 0; points = static_cast(intVal); if(fftSize > 0 && points > fftSize) { TrErrorAt(tr, line, col, "Value exceeds the overridden FFT size.\n"); return 0; } if(points < truncSize) { TrErrorAt(tr, line, col, "Value is below the truncation size.\n"); return 0; } hData->mIrPoints = points; if(fftSize <= 0) { hData->mFftSize = DEFAULT_FFTSIZE; hData->mIrSize = 1 + (DEFAULT_FFTSIZE / 2); } else { hData->mFftSize = fftSize; hData->mIrSize = 1 + (fftSize / 2); if(points > hData->mIrSize) hData->mIrSize = points; } hasPoints = 1; } else if(strcasecmp(ident, "radius") == 0) { if(hasRadius) { TrErrorAt(tr, line, col, "Redefinition of 'radius'.\n"); return 0; } if(!TrReadOperator(tr, "=")) return 0; if(!TrReadFloat(tr, MIN_RADIUS, MAX_RADIUS, &fpVal)) return 0; hData->mRadius = fpVal; hasRadius = 1; } else if(strcasecmp(ident, "distance") == 0) { uint count = 0; if(hasDistance) { TrErrorAt(tr, line, col, "Redefinition of 'distance'.\n"); return 0; } if(!TrReadOperator(tr, "=")) return 0; for(;;) { if(!TrReadFloat(tr, MIN_DISTANCE, MAX_DISTANCE, &fpVal)) return 0; if(count > 0 && fpVal <= distances[count - 1]) { TrError(tr, "Distances are not ascending.\n"); return 0; } distances[count++] = fpVal; if(!TrIsOperator(tr, ",")) break; if(count >= MAX_FD_COUNT) { TrError(tr, "Exceeded the maximum of %d fields.\n", MAX_FD_COUNT); return 0; } TrReadOperator(tr, ","); } if(fdCount != 0 && count != fdCount) { TrError(tr, "Did not match the specified number of %d fields.\n", fdCount); return 0; } fdCount = count; hasDistance = 1; } else if(strcasecmp(ident, "azimuths") == 0) { uint count = 0; if(hasAzimuths) { TrErrorAt(tr, line, col, "Redefinition of 'azimuths'.\n"); return 0; } if(!TrReadOperator(tr, "=")) return 0; evCounts[0] = 0; for(;;) { if(!TrReadInt(tr, MIN_AZ_COUNT, MAX_AZ_COUNT, &intVal)) return 0; azCounts[(count * MAX_EV_COUNT) + evCounts[count]++] = static_cast(intVal); if(TrIsOperator(tr, ",")) { if(evCounts[count] >= MAX_EV_COUNT) { TrError(tr, "Exceeded the maximum of %d elevations.\n", MAX_EV_COUNT); return 0; } TrReadOperator(tr, ","); } else { if(evCounts[count] < MIN_EV_COUNT) { TrErrorAt(tr, line, col, "Did not reach the minimum of %d azimuth counts.\n", MIN_EV_COUNT); return 0; } if(azCounts[count * MAX_EV_COUNT] != 1 || azCounts[(count * MAX_EV_COUNT) + evCounts[count] - 1] != 1) { TrError(tr, "Poles are not singular for field %d.\n", count - 1); return 0; } count++; if(!TrIsOperator(tr, ";")) break; if(count >= MAX_FD_COUNT) { TrError(tr, "Exceeded the maximum number of %d fields.\n", MAX_FD_COUNT); return 0; } evCounts[count] = 0; TrReadOperator(tr, ";"); } } if(fdCount != 0 && count != fdCount) { TrError(tr, "Did not match the specified number of %d fields.\n", fdCount); return 0; } fdCount = count; hasAzimuths = 1; } else { TrErrorAt(tr, line, col, "Expected a metric name.\n"); return 0; } TrSkipWhitespace(tr); } if(!(hasRate && hasPoints && hasRadius && hasDistance && hasAzimuths)) { TrErrorAt(tr, line, col, "Expected a metric name.\n"); return 0; } if(distances[0] < hData->mRadius) { TrError(tr, "Distance cannot start below head radius.\n"); return 0; } if(hData->mChannelType == CT_NONE) hData->mChannelType = CT_MONO; if(!PrepareHrirData(fdCount, distances, evCounts, azCounts.data(), hData)) { fprintf(stderr, "Error: Out of memory.\n"); exit(-1); } return 1; } // Parse an index triplet from the data set definition. static int ReadIndexTriplet(TokenReaderT *tr, const HrirDataT *hData, uint *fi, uint *ei, uint *ai) { int intVal; if(hData->mFdCount > 1) { if(!TrReadInt(tr, 0, static_cast(hData->mFdCount) - 1, &intVal)) return 0; *fi = static_cast(intVal); if(!TrReadOperator(tr, ",")) return 0; } else { *fi = 0; } if(!TrReadInt(tr, 0, static_cast(hData->mFds[*fi].mEvCount) - 1, &intVal)) return 0; *ei = static_cast(intVal); if(!TrReadOperator(tr, ",")) return 0; if(!TrReadInt(tr, 0, static_cast(hData->mFds[*fi].mEvs[*ei].mAzCount) - 1, &intVal)) return 0; *ai = static_cast(intVal); return 1; } // Match the source format from a given identifier. static SourceFormatT MatchSourceFormat(const char *ident) { if(strcasecmp(ident, "ascii") == 0) return SF_ASCII; if(strcasecmp(ident, "bin_le") == 0) return SF_BIN_LE; if(strcasecmp(ident, "bin_be") == 0) return SF_BIN_BE; if(strcasecmp(ident, "wave") == 0) return SF_WAVE; if(strcasecmp(ident, "sofa") == 0) return SF_SOFA; return SF_NONE; } // Match the source element type from a given identifier. static ElementTypeT MatchElementType(const char *ident) { if(strcasecmp(ident, "int") == 0) return ET_INT; if(strcasecmp(ident, "fp") == 0) return ET_FP; return ET_NONE; } // Parse and validate a source reference from the data set definition. static int ReadSourceRef(TokenReaderT *tr, SourceRefT *src) { char ident[MAX_IDENT_LEN+1]; uint line, col; double fpVal; int intVal; TrIndication(tr, &line, &col); if(!TrReadIdent(tr, MAX_IDENT_LEN, ident)) return 0; src->mFormat = MatchSourceFormat(ident); if(src->mFormat == SF_NONE) { TrErrorAt(tr, line, col, "Expected a source format.\n"); return 0; } if(!TrReadOperator(tr, "(")) return 0; if(src->mFormat == SF_SOFA) { if(!TrReadFloat(tr, MIN_DISTANCE, MAX_DISTANCE, &fpVal)) return 0; src->mRadius = fpVal; if(!TrReadOperator(tr, ",")) return 0; if(!TrReadFloat(tr, -90.0, 90.0, &fpVal)) return 0; src->mElevation = fpVal; if(!TrReadOperator(tr, ",")) return 0; if(!TrReadFloat(tr, -360.0, 360.0, &fpVal)) return 0; src->mAzimuth = fpVal; if(!TrReadOperator(tr, ":")) return 0; if(!TrReadInt(tr, 0, MAX_WAVE_CHANNELS, &intVal)) return 0; src->mType = ET_NONE; src->mSize = 0; src->mBits = 0; src->mChannel = (uint)intVal; src->mSkip = 0; } else if(src->mFormat == SF_WAVE) { if(!TrReadInt(tr, 0, MAX_WAVE_CHANNELS, &intVal)) return 0; src->mType = ET_NONE; src->mSize = 0; src->mBits = 0; src->mChannel = static_cast(intVal); src->mSkip = 0; } else { TrIndication(tr, &line, &col); if(!TrReadIdent(tr, MAX_IDENT_LEN, ident)) return 0; src->mType = MatchElementType(ident); if(src->mType == ET_NONE) { TrErrorAt(tr, line, col, "Expected a source element type.\n"); return 0; } if(src->mFormat == SF_BIN_LE || src->mFormat == SF_BIN_BE) { if(!TrReadOperator(tr, ",")) return 0; if(src->mType == ET_INT) { if(!TrReadInt(tr, MIN_BIN_SIZE, MAX_BIN_SIZE, &intVal)) return 0; src->mSize = static_cast(intVal); if(!TrIsOperator(tr, ",")) src->mBits = static_cast(8*src->mSize); else { TrReadOperator(tr, ","); TrIndication(tr, &line, &col); if(!TrReadInt(tr, -2147483647-1, 2147483647, &intVal)) return 0; if(std::abs(intVal) < MIN_BIN_BITS || static_cast(std::abs(intVal)) > (8*src->mSize)) { TrErrorAt(tr, line, col, "Expected a value of (+/-) %d to %d.\n", MIN_BIN_BITS, 8*src->mSize); return 0; } src->mBits = intVal; } } else { TrIndication(tr, &line, &col); if(!TrReadInt(tr, -2147483647-1, 2147483647, &intVal)) return 0; if(intVal != 4 && intVal != 8) { TrErrorAt(tr, line, col, "Expected a value of 4 or 8.\n"); return 0; } src->mSize = static_cast(intVal); src->mBits = 0; } } else if(src->mFormat == SF_ASCII && src->mType == ET_INT) { if(!TrReadOperator(tr, ",")) return 0; if(!TrReadInt(tr, MIN_ASCII_BITS, MAX_ASCII_BITS, &intVal)) return 0; src->mSize = 0; src->mBits = intVal; } else { src->mSize = 0; src->mBits = 0; } if(!TrIsOperator(tr, ";")) src->mSkip = 0; else { TrReadOperator(tr, ";"); if(!TrReadInt(tr, 0, 0x7FFFFFFF, &intVal)) return 0; src->mSkip = static_cast(intVal); } } if(!TrReadOperator(tr, ")")) return 0; if(TrIsOperator(tr, "@")) { TrReadOperator(tr, "@"); if(!TrReadInt(tr, 0, 0x7FFFFFFF, &intVal)) return 0; src->mOffset = static_cast(intVal); } else src->mOffset = 0; if(!TrReadOperator(tr, ":")) return 0; if(!TrReadString(tr, MAX_PATH_LEN, src->mPath)) return 0; return 1; } // Parse and validate a SOFA source reference from the data set definition. static int ReadSofaRef(TokenReaderT *tr, SourceRefT *src) { char ident[MAX_IDENT_LEN+1]; uint line, col; int intVal; TrIndication(tr, &line, &col); if(!TrReadIdent(tr, MAX_IDENT_LEN, ident)) return 0; src->mFormat = MatchSourceFormat(ident); if(src->mFormat != SF_SOFA) { TrErrorAt(tr, line, col, "Expected the SOFA source format.\n"); return 0; } src->mType = ET_NONE; src->mSize = 0; src->mBits = 0; src->mChannel = 0; src->mSkip = 0; if(TrIsOperator(tr, "@")) { TrReadOperator(tr, "@"); if(!TrReadInt(tr, 0, 0x7FFFFFFF, &intVal)) return 0; src->mOffset = (uint)intVal; } else src->mOffset = 0; if(!TrReadOperator(tr, ":")) return 0; if(!TrReadString(tr, MAX_PATH_LEN, src->mPath)) return 0; return 1; } // Match the target ear (index) from a given identifier. static int MatchTargetEar(const char *ident) { if(strcasecmp(ident, "left") == 0) return 0; if(strcasecmp(ident, "right") == 0) return 1; return -1; } // Process the list of sources in the data set definition. static int ProcessSources(const HeadModelT model, TokenReaderT *tr, HrirDataT *hData) { uint channels = (hData->mChannelType == CT_STEREO) ? 2 : 1; hData->mHrirsBase.resize(channels * hData->mIrCount * hData->mIrSize); double *hrirs = hData->mHrirsBase.data(); std::vector hrir(hData->mIrPoints); uint line, col, fi, ei, ai, ti; int count; printf("Loading sources..."); fflush(stdout); count = 0; while(TrIsOperator(tr, "[")) { double factor[2]{ 1.0, 1.0 }; TrIndication(tr, &line, &col); TrReadOperator(tr, "["); if(TrIsOperator(tr, "*")) { SourceRefT src; struct MYSOFA_EASY *sofa; uint si; TrReadOperator(tr, "*"); if(!TrReadOperator(tr, "]") || !TrReadOperator(tr, "=")) return 0; TrIndication(tr, &line, &col); if(!ReadSofaRef(tr, &src)) return 0; if(hData->mChannelType == CT_STEREO) { char type[MAX_IDENT_LEN+1]; ChannelTypeT channelType; if(!TrReadIdent(tr, MAX_IDENT_LEN, type)) return 0; channelType = MatchChannelType(type); switch(channelType) { case CT_NONE: TrErrorAt(tr, line, col, "Expected a channel type.\n"); return 0; case CT_MONO: src.mChannel = 0; break; case CT_STEREO: src.mChannel = 1; break; } } else { char type[MAX_IDENT_LEN+1]; ChannelTypeT channelType; if(!TrReadIdent(tr, MAX_IDENT_LEN, type)) return 0; channelType = MatchChannelType(type); if(channelType != CT_MONO) { TrErrorAt(tr, line, col, "Expected a mono channel type.\n"); return 0; } src.mChannel = 0; } sofa = LoadSofaFile(&src, hData->mIrRate, hData->mIrPoints); if(!sofa) return 0; for(si = 0;si < sofa->hrtf->M;si++) { printf("\rLoading sources... %d of %d", si+1, sofa->hrtf->M); fflush(stdout); float aer[3] = { sofa->hrtf->SourcePosition.values[3*si], sofa->hrtf->SourcePosition.values[3*si + 1], sofa->hrtf->SourcePosition.values[3*si + 2] }; mysofa_c2s(aer); if(std::fabs(aer[1]) >= 89.999f) aer[0] = 0.0f; else aer[0] = std::fmod(360.0f - aer[0], 360.0f); for(fi = 0;fi < hData->mFdCount;fi++) { double delta = aer[2] - hData->mFds[fi].mDistance; if(std::abs(delta) < 0.001) break; } if(fi >= hData->mFdCount) continue; double ef{(90.0 + aer[1]) * (hData->mFds[fi].mEvCount - 1) / 180.0}; ei = (int)std::round(ef); ef = (ef - ei) * 180.0f / (hData->mFds[fi].mEvCount - 1); if(std::abs(ef) >= 0.1) continue; double af{aer[0] * hData->mFds[fi].mEvs[ei].mAzCount / 360.0f}; ai = (int)std::round(af); af = (af - ai) * 360.0f / hData->mFds[fi].mEvs[ei].mAzCount; ai = ai % hData->mFds[fi].mEvs[ei].mAzCount; if(std::abs(af) >= 0.1) continue; HrirAzT *azd = &hData->mFds[fi].mEvs[ei].mAzs[ai]; if(azd->mIrs[0] != nullptr) { TrErrorAt(tr, line, col, "Redefinition of source [ %d, %d, %d ].\n", fi, ei, ai); return 0; } ExtractSofaHrir(sofa, si, 0, src.mOffset, hData->mIrPoints, hrir.data()); azd->mIrs[0] = &hrirs[hData->mIrSize * azd->mIndex]; if(model == HM_DATASET) azd->mDelays[0] = AverageHrirOnset(hData->mIrRate, hData->mIrPoints, hrir.data(), 1.0, azd->mDelays[0]); AverageHrirMagnitude(hData->mIrPoints, hData->mFftSize, hrir.data(), 1.0, azd->mIrs[0]); if(src.mChannel == 1) { ExtractSofaHrir(sofa, si, 1, src.mOffset, hData->mIrPoints, hrir.data()); azd->mIrs[1] = &hrirs[hData->mIrSize * (hData->mIrCount + azd->mIndex)]; if(model == HM_DATASET) azd->mDelays[1] = AverageHrirOnset(hData->mIrRate, hData->mIrPoints, hrir.data(), 1.0, azd->mDelays[1]); AverageHrirMagnitude(hData->mIrPoints, hData->mFftSize, hrir.data(), 1.0, azd->mIrs[1]); } // TODO: Since some SOFA files contain minimum phase HRIRs, // it would be beneficial to check for per-measurement delays // (when available) to reconstruct the HRTDs. } continue; } if(!ReadIndexTriplet(tr, hData, &fi, &ei, &ai)) return 0; if(!TrReadOperator(tr, "]")) return 0; HrirAzT *azd = &hData->mFds[fi].mEvs[ei].mAzs[ai]; if(azd->mIrs[0] != nullptr) { TrErrorAt(tr, line, col, "Redefinition of source.\n"); return 0; } if(!TrReadOperator(tr, "=")) return 0; for(;;) { SourceRefT src; uint ti = 0; if(!ReadSourceRef(tr, &src)) return 0; // TODO: Would be nice to display 'x of y files', but that would // require preparing the source refs first to get a total count // before loading them. ++count; printf("\rLoading sources... %d file%s", count, (count==1)?"":"s"); fflush(stdout); if(!LoadSource(&src, hData->mIrRate, hData->mIrPoints, hrir.data())) return 0; if(hData->mChannelType == CT_STEREO) { char ident[MAX_IDENT_LEN+1]; if(!TrReadIdent(tr, MAX_IDENT_LEN, ident)) return 0; ti = MatchTargetEar(ident); if(static_cast(ti) < 0) { TrErrorAt(tr, line, col, "Expected a target ear.\n"); return 0; } } azd->mIrs[ti] = &hrirs[hData->mIrSize * (ti * hData->mIrCount + azd->mIndex)]; if(model == HM_DATASET) azd->mDelays[ti] = AverageHrirOnset(hData->mIrRate, hData->mIrPoints, hrir.data(), 1.0 / factor[ti], azd->mDelays[ti]); AverageHrirMagnitude(hData->mIrPoints, hData->mFftSize, hrir.data(), 1.0 / factor[ti], azd->mIrs[ti]); factor[ti] += 1.0; if(!TrIsOperator(tr, "+")) break; TrReadOperator(tr, "+"); } if(hData->mChannelType == CT_STEREO) { if(azd->mIrs[0] == nullptr) { TrErrorAt(tr, line, col, "Missing left ear source reference(s).\n"); return 0; } else if(azd->mIrs[1] == nullptr) { TrErrorAt(tr, line, col, "Missing right ear source reference(s).\n"); return 0; } } } printf("\n"); for(fi = 0;fi < hData->mFdCount;fi++) { for(ei = 0;ei < hData->mFds[fi].mEvCount;ei++) { for(ai = 0;ai < hData->mFds[fi].mEvs[ei].mAzCount;ai++) { HrirAzT *azd = &hData->mFds[fi].mEvs[ei].mAzs[ai]; if(azd->mIrs[0] != nullptr) break; } if(ai < hData->mFds[fi].mEvs[ei].mAzCount) break; } if(ei >= hData->mFds[fi].mEvCount) { TrError(tr, "Missing source references [ %d, *, * ].\n", fi); return 0; } hData->mFds[fi].mEvStart = ei; for(;ei < hData->mFds[fi].mEvCount;ei++) { for(ai = 0;ai < hData->mFds[fi].mEvs[ei].mAzCount;ai++) { HrirAzT *azd = &hData->mFds[fi].mEvs[ei].mAzs[ai]; if(azd->mIrs[0] == nullptr) { TrError(tr, "Missing source reference [ %d, %d, %d ].\n", fi, ei, ai); return 0; } } } } for(ti = 0;ti < channels;ti++) { for(fi = 0;fi < hData->mFdCount;fi++) { for(ei = 0;ei < hData->mFds[fi].mEvCount;ei++) { for(ai = 0;ai < hData->mFds[fi].mEvs[ei].mAzCount;ai++) { HrirAzT *azd = &hData->mFds[fi].mEvs[ei].mAzs[ai]; azd->mIrs[ti] = &hrirs[hData->mIrSize * (ti * hData->mIrCount + azd->mIndex)]; } } } } if(!TrLoad(tr)) { mysofa_cache_release_all(); return 1; } TrError(tr, "Errant data at end of source list.\n"); mysofa_cache_release_all(); return 0; } /* Parse the data set definition and process the source data, storing the * resulting data set as desired. If the input name is NULL it will read * from standard input. */ static int ProcessDefinition(const char *inName, const uint outRate, const uint fftSize, const int equalize, const int surface, const double limit, const uint truncSize, const HeadModelT model, const double radius, const char *outName) { char rateStr[8+1], expName[MAX_PATH_LEN]; TokenReaderT tr; HrirDataT hData; FILE *fp; int ret; fprintf(stdout, "Reading HRIR definition from %s...\n", inName?inName:"stdin"); if(inName != nullptr) { fp = fopen(inName, "r"); if(fp == nullptr) { fprintf(stderr, "\nError: Could not open definition file '%s'\n", inName); return 0; } TrSetup(fp, inName, &tr); } else { fp = stdin; TrSetup(fp, "", &tr); } if(!ProcessMetrics(&tr, fftSize, truncSize, &hData)) { if(inName != nullptr) fclose(fp); return 0; } if(!ProcessSources(model, &tr, &hData)) { if(inName) fclose(fp); return 0; } if(fp != stdin) fclose(fp); if(equalize) { uint c = (hData.mChannelType == CT_STEREO) ? 2 : 1; uint m = 1 + hData.mFftSize / 2; std::vector dfa(c * m); if(hData.mFdCount > 1) { fprintf(stdout, "Balancing field magnitudes...\n"); BalanceFieldMagnitudes(&hData, c, m); } fprintf(stdout, "Calculating diffuse-field average...\n"); CalculateDiffuseFieldAverage(&hData, c, m, surface, limit, dfa.data()); fprintf(stdout, "Performing diffuse-field equalization...\n"); DiffuseFieldEqualize(c, m, dfa.data(), &hData); } fprintf(stdout, "Performing minimum phase reconstruction...\n"); ReconstructHrirs(&hData); if(outRate != 0 && outRate != hData.mIrRate) { fprintf(stdout, "Resampling HRIRs...\n"); ResampleHrirs(outRate, &hData); } fprintf(stdout, "Truncating minimum-phase HRIRs...\n"); hData.mIrPoints = truncSize; fprintf(stdout, "Synthesizing missing elevations...\n"); if(model == HM_DATASET) SynthesizeOnsets(&hData); SynthesizeHrirs(&hData); fprintf(stdout, "Normalizing final HRIRs...\n"); NormalizeHrirs(&hData); fprintf(stdout, "Calculating impulse delays...\n"); CalculateHrtds(model, (radius > DEFAULT_CUSTOM_RADIUS) ? radius : hData.mRadius, &hData); snprintf(rateStr, 8, "%u", hData.mIrRate); StrSubst(outName, "%r", rateStr, MAX_PATH_LEN, expName); fprintf(stdout, "Creating MHR data set %s...\n", expName); ret = StoreMhr(&hData, expName); return ret; } static void PrintHelp(const char *argv0, FILE *ofile) { fprintf(ofile, "Usage: %s [