| 1 | /* |
| 2 | * Copyright (C) 2010 Google Inc. All rights reserved. |
| 3 | * |
| 4 | * Redistribution and use in source and binary forms, with or without |
| 5 | * modification, are permitted provided that the following conditions |
| 6 | * are met: |
| 7 | * |
| 8 | * 1. Redistributions of source code must retain the above copyright |
| 9 | * notice, this list of conditions and the following disclaimer. |
| 10 | * 2. Redistributions in binary form must reproduce the above copyright |
| 11 | * notice, this list of conditions and the following disclaimer in the |
| 12 | * documentation and/or other materials provided with the distribution. |
| 13 | * 3. Neither the name of Apple Inc. ("Apple") nor the names of |
| 14 | * its contributors may be used to endorse or promote products derived |
| 15 | * from this software without specific prior written permission. |
| 16 | * |
| 17 | * THIS SOFTWARE IS PROVIDED BY APPLE AND ITS CONTRIBUTORS "AS IS" AND ANY |
| 18 | * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED |
| 19 | * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE |
| 20 | * DISCLAIMED. IN NO EVENT SHALL APPLE OR ITS CONTRIBUTORS BE LIABLE FOR ANY |
| 21 | * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES |
| 22 | * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; |
| 23 | * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND |
| 24 | * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT |
| 25 | * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF |
| 26 | * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
| 27 | */ |
| 28 | |
| 29 | #include "config.h" |
| 30 | |
| 31 | #if ENABLE(WEB_AUDIO) |
| 32 | |
| 33 | #include "HRTFElevation.h" |
| 34 | |
| 35 | #include "AudioBus.h" |
| 36 | #include "AudioFileReader.h" |
| 37 | #include "Biquad.h" |
| 38 | #include "FFTFrame.h" |
| 39 | #include "HRTFDatabaseLoader.h" |
| 40 | #include "HRTFPanner.h" |
| 41 | #include <algorithm> |
| 42 | #include <math.h> |
| 43 | #include <wtf/NeverDestroyed.h> |
| 44 | |
| 45 | namespace WebCore { |
| 46 | |
| 47 | const unsigned HRTFElevation::AzimuthSpacing = 15; |
| 48 | const unsigned HRTFElevation::NumberOfRawAzimuths = 360 / AzimuthSpacing; |
| 49 | const unsigned HRTFElevation::InterpolationFactor = 8; |
| 50 | const unsigned HRTFElevation::NumberOfTotalAzimuths = NumberOfRawAzimuths * InterpolationFactor; |
| 51 | |
| 52 | // Total number of components of an HRTF database. |
| 53 | const size_t TotalNumberOfResponses = 240; |
| 54 | |
| 55 | // Number of frames in an individual impulse response. |
| 56 | const size_t ResponseFrameSize = 256; |
| 57 | |
| 58 | // Sample-rate of the spatialization impulse responses as stored in the resource file. |
| 59 | // The impulse responses may be resampled to a different sample-rate (depending on the audio hardware) when they are loaded. |
| 60 | const float ResponseSampleRate = 44100; |
| 61 | |
| 62 | #if PLATFORM(COCOA) || USE(WEBAUDIO_GSTREAMER) |
| 63 | #define USE_CONCATENATED_IMPULSE_RESPONSES |
| 64 | #endif |
| 65 | |
| 66 | #ifdef USE_CONCATENATED_IMPULSE_RESPONSES |
| 67 | // Lazily load a concatenated HRTF database for given subject and store it in a |
| 68 | // local hash table to ensure quick efficient future retrievals. |
| 69 | static AudioBus* getConcatenatedImpulseResponsesForSubject(const String& subjectName) |
| 70 | { |
| 71 | typedef HashMap<String, AudioBus*> AudioBusMap; |
| 72 | static NeverDestroyed<AudioBusMap> audioBusMap; |
| 73 | |
| 74 | AudioBus* bus; |
| 75 | AudioBusMap::iterator iterator = audioBusMap.get().find(subjectName); |
| 76 | if (iterator == audioBusMap.get().end()) { |
| 77 | auto concatenatedImpulseResponses = AudioBus::loadPlatformResource(subjectName.utf8().data(), ResponseSampleRate); |
| 78 | ASSERT(concatenatedImpulseResponses); |
| 79 | if (!concatenatedImpulseResponses) |
| 80 | return 0; |
| 81 | |
| 82 | bus = concatenatedImpulseResponses.leakRef(); |
| 83 | audioBusMap.get().set(subjectName, bus); |
| 84 | } else |
| 85 | bus = iterator->value; |
| 86 | |
| 87 | size_t responseLength = bus->length(); |
| 88 | size_t expectedLength = static_cast<size_t>(TotalNumberOfResponses * ResponseFrameSize); |
| 89 | |
| 90 | // Check number of channels and length. For now these are fixed and known. |
| 91 | bool isBusGood = responseLength == expectedLength && bus->numberOfChannels() == 2; |
| 92 | ASSERT(isBusGood); |
| 93 | if (!isBusGood) |
| 94 | return 0; |
| 95 | |
| 96 | return bus; |
| 97 | } |
| 98 | #endif |
| 99 | |
| 100 | // Takes advantage of the symmetry and creates a composite version of the two measured versions. For example, we have both azimuth 30 and -30 degrees |
| 101 | // where the roles of left and right ears are reversed with respect to each other. |
| 102 | bool HRTFElevation::calculateSymmetricKernelsForAzimuthElevation(int azimuth, int elevation, float sampleRate, const String& subjectName, |
| 103 | RefPtr<HRTFKernel>& kernelL, RefPtr<HRTFKernel>& kernelR) |
| 104 | { |
| 105 | RefPtr<HRTFKernel> kernelL1; |
| 106 | RefPtr<HRTFKernel> kernelR1; |
| 107 | bool success = calculateKernelsForAzimuthElevation(azimuth, elevation, sampleRate, subjectName, kernelL1, kernelR1); |
| 108 | if (!success) |
| 109 | return false; |
| 110 | |
| 111 | // And symmetric version |
| 112 | int symmetricAzimuth = !azimuth ? 0 : 360 - azimuth; |
| 113 | |
| 114 | RefPtr<HRTFKernel> kernelL2; |
| 115 | RefPtr<HRTFKernel> kernelR2; |
| 116 | success = calculateKernelsForAzimuthElevation(symmetricAzimuth, elevation, sampleRate, subjectName, kernelL2, kernelR2); |
| 117 | if (!success) |
| 118 | return false; |
| 119 | |
| 120 | // Notice L/R reversal in symmetric version. |
| 121 | kernelL = HRTFKernel::createInterpolatedKernel(kernelL1.get(), kernelR2.get(), 0.5f); |
| 122 | kernelR = HRTFKernel::createInterpolatedKernel(kernelR1.get(), kernelL2.get(), 0.5f); |
| 123 | |
| 124 | return true; |
| 125 | } |
| 126 | |
| 127 | bool HRTFElevation::calculateKernelsForAzimuthElevation(int azimuth, int elevation, float sampleRate, const String& subjectName, |
| 128 | RefPtr<HRTFKernel>& kernelL, RefPtr<HRTFKernel>& kernelR) |
| 129 | { |
| 130 | // Valid values for azimuth are 0 -> 345 in 15 degree increments. |
| 131 | // Valid values for elevation are -45 -> +90 in 15 degree increments. |
| 132 | |
| 133 | bool isAzimuthGood = azimuth >= 0 && azimuth <= 345 && (azimuth / 15) * 15 == azimuth; |
| 134 | ASSERT(isAzimuthGood); |
| 135 | if (!isAzimuthGood) |
| 136 | return false; |
| 137 | |
| 138 | bool isElevationGood = elevation >= -45 && elevation <= 90 && (elevation / 15) * 15 == elevation; |
| 139 | ASSERT(isElevationGood); |
| 140 | if (!isElevationGood) |
| 141 | return false; |
| 142 | |
| 143 | // Construct the resource name from the subject name, azimuth, and elevation, for example: |
| 144 | // "IRC_Composite_C_R0195_T015_P000" |
| 145 | // Note: the passed in subjectName is not a string passed in via JavaScript or the web. |
| 146 | // It's passed in as an internal ASCII identifier and is an implementation detail. |
| 147 | int positiveElevation = elevation < 0 ? elevation + 360 : elevation; |
| 148 | |
| 149 | #ifdef USE_CONCATENATED_IMPULSE_RESPONSES |
| 150 | AudioBus* bus(getConcatenatedImpulseResponsesForSubject(subjectName)); |
| 151 | |
| 152 | if (!bus) |
| 153 | return false; |
| 154 | |
| 155 | int elevationIndex = positiveElevation / AzimuthSpacing; |
| 156 | if (positiveElevation > 90) |
| 157 | elevationIndex -= AzimuthSpacing; |
| 158 | |
| 159 | // The concatenated impulse response is a bus containing all |
| 160 | // the elevations per azimuth, for all azimuths by increasing |
| 161 | // order. So for a given azimuth and elevation we need to compute |
| 162 | // the index of the wanted audio frames in the concatenated table. |
| 163 | unsigned index = ((azimuth / AzimuthSpacing) * HRTFDatabase::NumberOfRawElevations) + elevationIndex; |
| 164 | bool isIndexGood = index < TotalNumberOfResponses; |
| 165 | ASSERT(isIndexGood); |
| 166 | if (!isIndexGood) |
| 167 | return false; |
| 168 | |
| 169 | // Extract the individual impulse response from the concatenated |
| 170 | // responses and potentially sample-rate convert it to the desired |
| 171 | // (hardware) sample-rate. |
| 172 | unsigned startFrame = index * ResponseFrameSize; |
| 173 | unsigned stopFrame = startFrame + ResponseFrameSize; |
| 174 | auto preSampleRateConvertedResponse = AudioBus::createBufferFromRange(bus, startFrame, stopFrame); |
| 175 | auto response = AudioBus::createBySampleRateConverting(preSampleRateConvertedResponse.get(), false, sampleRate); |
| 176 | AudioChannel* leftEarImpulseResponse = response->channel(AudioBus::ChannelLeft); |
| 177 | AudioChannel* rightEarImpulseResponse = response->channel(AudioBus::ChannelRight); |
| 178 | #else |
| 179 | String resourceName = makeString("IRC_" , subjectName, "_C_R0195_T" , pad('0', 3, azimuth), "_P" , pad('0', 3, positiveElevation)); |
| 180 | |
| 181 | RefPtr<AudioBus> impulseResponse(AudioBus::loadPlatformResource(resourceName.utf8().data(), sampleRate)); |
| 182 | |
| 183 | ASSERT(impulseResponse.get()); |
| 184 | if (!impulseResponse.get()) |
| 185 | return false; |
| 186 | |
| 187 | size_t responseLength = impulseResponse->length(); |
| 188 | size_t expectedLength = static_cast<size_t>(256 * (sampleRate / 44100.0)); |
| 189 | |
| 190 | // Check number of channels and length. For now these are fixed and known. |
| 191 | bool isBusGood = responseLength == expectedLength && impulseResponse->numberOfChannels() == 2; |
| 192 | ASSERT(isBusGood); |
| 193 | if (!isBusGood) |
| 194 | return false; |
| 195 | |
| 196 | AudioChannel* leftEarImpulseResponse = impulseResponse->channelByType(AudioBus::ChannelLeft); |
| 197 | AudioChannel* rightEarImpulseResponse = impulseResponse->channelByType(AudioBus::ChannelRight); |
| 198 | #endif |
| 199 | |
| 200 | // Note that depending on the fftSize returned by the panner, we may be truncating the impulse response we just loaded in. |
| 201 | const size_t fftSize = HRTFPanner::fftSizeForSampleRate(sampleRate); |
| 202 | kernelL = HRTFKernel::create(leftEarImpulseResponse, fftSize, sampleRate); |
| 203 | kernelR = HRTFKernel::create(rightEarImpulseResponse, fftSize, sampleRate); |
| 204 | |
| 205 | return true; |
| 206 | } |
| 207 | |
| 208 | // The range of elevations for the IRCAM impulse responses varies depending on azimuth, but the minimum elevation appears to always be -45. |
| 209 | // |
| 210 | // Here's how it goes: |
| 211 | static const int maxElevations[] = { |
| 212 | // Azimuth |
| 213 | // |
| 214 | 90, // 0 |
| 215 | 45, // 15 |
| 216 | 60, // 30 |
| 217 | 45, // 45 |
| 218 | 75, // 60 |
| 219 | 45, // 75 |
| 220 | 60, // 90 |
| 221 | 45, // 105 |
| 222 | 75, // 120 |
| 223 | 45, // 135 |
| 224 | 60, // 150 |
| 225 | 45, // 165 |
| 226 | 75, // 180 |
| 227 | 45, // 195 |
| 228 | 60, // 210 |
| 229 | 45, // 225 |
| 230 | 75, // 240 |
| 231 | 45, // 255 |
| 232 | 60, // 270 |
| 233 | 45, // 285 |
| 234 | 75, // 300 |
| 235 | 45, // 315 |
| 236 | 60, // 330 |
| 237 | 45 // 345 |
| 238 | }; |
| 239 | |
| 240 | std::unique_ptr<HRTFElevation> HRTFElevation::createForSubject(const String& subjectName, int elevation, float sampleRate) |
| 241 | { |
| 242 | bool isElevationGood = elevation >= -45 && elevation <= 90 && (elevation / 15) * 15 == elevation; |
| 243 | ASSERT(isElevationGood); |
| 244 | if (!isElevationGood) |
| 245 | return nullptr; |
| 246 | |
| 247 | auto kernelListL = std::make_unique<HRTFKernelList>(NumberOfTotalAzimuths); |
| 248 | auto kernelListR = std::make_unique<HRTFKernelList>(NumberOfTotalAzimuths); |
| 249 | |
| 250 | // Load convolution kernels from HRTF files. |
| 251 | int interpolatedIndex = 0; |
| 252 | for (unsigned rawIndex = 0; rawIndex < NumberOfRawAzimuths; ++rawIndex) { |
| 253 | // Don't let elevation exceed maximum for this azimuth. |
| 254 | int maxElevation = maxElevations[rawIndex]; |
| 255 | int actualElevation = std::min(elevation, maxElevation); |
| 256 | |
| 257 | bool success = calculateKernelsForAzimuthElevation(rawIndex * AzimuthSpacing, actualElevation, sampleRate, subjectName, kernelListL->at(interpolatedIndex), kernelListR->at(interpolatedIndex)); |
| 258 | if (!success) |
| 259 | return nullptr; |
| 260 | |
| 261 | interpolatedIndex += InterpolationFactor; |
| 262 | } |
| 263 | |
| 264 | // Now go back and interpolate intermediate azimuth values. |
| 265 | for (unsigned i = 0; i < NumberOfTotalAzimuths; i += InterpolationFactor) { |
| 266 | int j = (i + InterpolationFactor) % NumberOfTotalAzimuths; |
| 267 | |
| 268 | // Create the interpolated convolution kernels and delays. |
| 269 | for (unsigned jj = 1; jj < InterpolationFactor; ++jj) { |
| 270 | float x = float(jj) / float(InterpolationFactor); // interpolate from 0 -> 1 |
| 271 | |
| 272 | (*kernelListL)[i + jj] = HRTFKernel::createInterpolatedKernel(kernelListL->at(i).get(), kernelListL->at(j).get(), x); |
| 273 | (*kernelListR)[i + jj] = HRTFKernel::createInterpolatedKernel(kernelListR->at(i).get(), kernelListR->at(j).get(), x); |
| 274 | } |
| 275 | } |
| 276 | |
| 277 | return std::make_unique<HRTFElevation>(WTFMove(kernelListL), WTFMove(kernelListR), elevation, sampleRate); |
| 278 | } |
| 279 | |
| 280 | std::unique_ptr<HRTFElevation> HRTFElevation::createByInterpolatingSlices(HRTFElevation* hrtfElevation1, HRTFElevation* hrtfElevation2, float x, float sampleRate) |
| 281 | { |
| 282 | ASSERT(hrtfElevation1 && hrtfElevation2); |
| 283 | if (!hrtfElevation1 || !hrtfElevation2) |
| 284 | return nullptr; |
| 285 | |
| 286 | ASSERT(x >= 0.0 && x < 1.0); |
| 287 | |
| 288 | auto kernelListL = std::make_unique<HRTFKernelList>(NumberOfTotalAzimuths); |
| 289 | auto kernelListR = std::make_unique<HRTFKernelList>(NumberOfTotalAzimuths); |
| 290 | |
| 291 | HRTFKernelList* kernelListL1 = hrtfElevation1->kernelListL(); |
| 292 | HRTFKernelList* kernelListR1 = hrtfElevation1->kernelListR(); |
| 293 | HRTFKernelList* kernelListL2 = hrtfElevation2->kernelListL(); |
| 294 | HRTFKernelList* kernelListR2 = hrtfElevation2->kernelListR(); |
| 295 | |
| 296 | // Interpolate kernels of corresponding azimuths of the two elevations. |
| 297 | for (unsigned i = 0; i < NumberOfTotalAzimuths; ++i) { |
| 298 | (*kernelListL)[i] = HRTFKernel::createInterpolatedKernel(kernelListL1->at(i).get(), kernelListL2->at(i).get(), x); |
| 299 | (*kernelListR)[i] = HRTFKernel::createInterpolatedKernel(kernelListR1->at(i).get(), kernelListR2->at(i).get(), x); |
| 300 | } |
| 301 | |
| 302 | // Interpolate elevation angle. |
| 303 | double angle = (1.0 - x) * hrtfElevation1->elevationAngle() + x * hrtfElevation2->elevationAngle(); |
| 304 | |
| 305 | return std::make_unique<HRTFElevation>(WTFMove(kernelListL), WTFMove(kernelListR), static_cast<int>(angle), sampleRate); |
| 306 | } |
| 307 | |
| 308 | void HRTFElevation::getKernelsFromAzimuth(double azimuthBlend, unsigned azimuthIndex, HRTFKernel* &kernelL, HRTFKernel* &kernelR, double& frameDelayL, double& frameDelayR) |
| 309 | { |
| 310 | bool checkAzimuthBlend = azimuthBlend >= 0.0 && azimuthBlend < 1.0; |
| 311 | ASSERT(checkAzimuthBlend); |
| 312 | if (!checkAzimuthBlend) |
| 313 | azimuthBlend = 0.0; |
| 314 | |
| 315 | unsigned numKernels = m_kernelListL->size(); |
| 316 | |
| 317 | bool isIndexGood = azimuthIndex < numKernels; |
| 318 | ASSERT(isIndexGood); |
| 319 | if (!isIndexGood) { |
| 320 | kernelL = 0; |
| 321 | kernelR = 0; |
| 322 | return; |
| 323 | } |
| 324 | |
| 325 | // Return the left and right kernels. |
| 326 | kernelL = m_kernelListL->at(azimuthIndex).get(); |
| 327 | kernelR = m_kernelListR->at(azimuthIndex).get(); |
| 328 | |
| 329 | frameDelayL = m_kernelListL->at(azimuthIndex)->frameDelay(); |
| 330 | frameDelayR = m_kernelListR->at(azimuthIndex)->frameDelay(); |
| 331 | |
| 332 | int azimuthIndex2 = (azimuthIndex + 1) % numKernels; |
| 333 | double frameDelay2L = m_kernelListL->at(azimuthIndex2)->frameDelay(); |
| 334 | double frameDelay2R = m_kernelListR->at(azimuthIndex2)->frameDelay(); |
| 335 | |
| 336 | // Linearly interpolate delays. |
| 337 | frameDelayL = (1.0 - azimuthBlend) * frameDelayL + azimuthBlend * frameDelay2L; |
| 338 | frameDelayR = (1.0 - azimuthBlend) * frameDelayR + azimuthBlend * frameDelay2R; |
| 339 | } |
| 340 | |
| 341 | } // namespace WebCore |
| 342 | |
| 343 | #endif // ENABLE(WEB_AUDIO) |
| 344 | |