Spaces:
Running
Running
File size: 17,317 Bytes
5f923cd | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 | // Copyright 2025 The ODML Authors.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
#ifndef THIRD_PARTY_ODML_LITERT_LM_RUNTIME_COMPONENTS_PREPROCESSOR_AUDIO_PREPROCESSOR_H_
#define THIRD_PARTY_ODML_LITERT_LM_RUNTIME_COMPONENTS_PREPROCESSOR_AUDIO_PREPROCESSOR_H_
#include <array>
#include <ostream>
#include "absl/status/statusor.h" // from @com_google_absl
#include "runtime/engine/io_types.h"
#include "runtime/util/status_macros.h" // IWYU pragma: keep
namespace litert::lm {
// Configuration for audio preprocessing.
class AudioPreprocessorConfig {
public:
// The padding type of for FFT bins.
enum FftPaddingType {
// Right padding. The resulted FFT frame will be right padding with zeros or
// truncated to the given FFT frame length.
kRight = 0,
// Center padding. The results FFT frame will be left and right padding with
// zeros with same amount, or truncated with same amount on left and right,
// to the given FFT frame length.
kCenter = 1
};
// Creates Google's Universal Speech Model (USM) audio preprocessing
// configuration.
static AudioPreprocessorConfig CreateDefaultUsmConfig() {
return AudioPreprocessorConfig(
/* sample_rate_hz= */ 16000,
/* num_channels= */ 1,
/* frame_length= */ 512,
/* hop_length= */ 160,
/* fft_length = */ 1024,
/* input_scale = */ 32768,
/* pre_emphasis_factor = */ 0.97,
/* num_mel_bins= */ 128,
/* mel_low_hz= */ 125.0,
/* mel_high_hz= */ 7500.0,
/* mel_floor= */ 1e-6,
/* normalize_mel= */ true,
/* add_floor_to_mel_before_log= */ false,
/* semicausal_padding= */ false, /* non_zero_hanning= */ true,
/* periodic_hanning= */ true,
/* fft_padding_type= */ FftPaddingType::kRight);
}
static AudioPreprocessorConfig Create(
int sample_rate_hz, int num_channels, int frame_length, int hop_length,
int fft_length, float input_scale, float pre_emphasis_factor,
int num_mel_bins, float mel_low_hz, float mel_high_hz, float mel_floor,
bool normalize_mel, bool add_floor_to_mel_before_log,
bool semicausal_padding, bool non_zero_hanning, bool periodic_hanning,
FftPaddingType fft_padding_type) {
return AudioPreprocessorConfig(
sample_rate_hz, num_channels, frame_length, hop_length, fft_length,
input_scale, pre_emphasis_factor, num_mel_bins, mel_low_hz, mel_high_hz,
mel_floor, normalize_mel, add_floor_to_mel_before_log,
semicausal_padding, non_zero_hanning, periodic_hanning,
fft_padding_type);
}
friend std::ostream& operator<<(std::ostream& os,
const FftPaddingType& padding_type) {
switch (padding_type) {
case FftPaddingType::kRight:
os << "right";
break;
case FftPaddingType::kCenter:
os << "center";
break;
default:
os << "unknown";
break;
}
return os;
}
// Allows logging of the config.
friend std::ostream& operator<<(std::ostream& os,
const AudioPreprocessorConfig& config) {
os << "AudioPreprocessorConfig {\n";
os << " sample_rate_hz: " << config.GetSampleRateHz() << "\n";
os << " num_channels: " << config.GetNumChannels() << "\n";
os << " input_scale: " << config.GetInputScale() << "\n";
os << " pre_emphasis_factor: " << config.GetPreEmphasisFactor() << "\n";
os << " fft_length: " << config.GetFftLength() << "\n";
os << " fft_bins: " << config.GetFftBins() << "\n";
os << " frame_length: " << config.GetFrameLength() << "\n";
os << " hop_length: " << config.GetHopLength() << "\n";
os << " num_mel_bins: " << config.GetNumMelBins() << "\n";
os << " mel_low_hz: " << config.GetMelLowHz() << "\n";
os << " mel_high_hz: " << config.GetMelHighHz() << "\n";
os << " mel_floor: " << config.GetMelFloor() << "\n";
os << " normalize_mel: " << config.GetNormalizeMel() << "\n";
os << " add_floor_to_mel_before_log: "
<< config.GetAddFloorToMelBeforeLog() << "\n";
os << " semicausal_padding: " << config.GetSemicausalPadding() << "\n";
os << " non_zero_hanning: " << config.GetNonZeroHanning() << "\n";
os << " periodic_hanning: " << config.GetPeriodicHanning() << "\n";
os << " fft_padding_type: " << config.GetFftPaddingType() << "\n";
os << "}";
return os;
}
// Getter APIs.
// The sample rate while loading the audio. The audio should be resampled to
// the configured sample rate.
int GetSampleRateHz() const { return sample_rate_hz_; }
// The number of audio channels the preprocessor expect from the audio
// content.
int GetNumChannels() const { return num_channels_; }
// The scale applied to the audio PCM frames before processing to
// spectrogram.
float GetInputScale() const { return input_scale_; }
// The pre-emphasis factor applied to the audio before processing to
// spectrogram.
float GetPreEmphasisFactor() const { return pre_emphasis_factor_; }
// The FFT length used for processing the audio.
int GetFftLength() const { return fft_length_; }
// The number of FFT bins used for real-sequence Fourier transform (RFFT) and
// Mel spectrogram processing. It is derived from the FFT length as FFT
// length / 2 + 1.
int GetFftBins() const { return fft_bins_; }
// The frame length used for for each frame of Short-Time Fourier Transform
// (STFT).
int GetFrameLength() const { return frame_length_; }
// The hop length used for in sliding window of Short-Time Fourier Transform
// (STFT).
int GetHopLength() const { return hop_length_; }
// The number of Mel bins used for Mel spectrogram processing.
int GetNumMelBins() const { return num_mel_bins_; }
// The lower bound of the Mel frequency range.
float GetMelLowHz() const { return mel_low_hz_; }
// The upper bound of the Mel frequency range.
float GetMelHighHz() const { return mel_high_hz_; }
// The floor value of the Mel spectrogram.
float GetMelFloor() const { return mel_floor_; }
// Whether to normalize the Mel spectrogram with precalculated mean and std
// dev.
bool GetNormalizeMel() const { return normalize_mel_; }
// Whether to add the floor value to the Mel spectrogram before taking the
// logarithm.
bool GetAddFloorToMelBeforeLog() const {
return add_floor_to_mel_before_log_;
}
// Whether to use semicausal padding for the audio frames.
bool GetSemicausalPadding() const { return semicausal_padding_; }
// Whether to use non-zero Hanning window for FFT.
bool GetNonZeroHanning() const { return non_zero_hanning_; }
// Whether to use the periodic Hanning window for FFT.
bool GetPeriodicHanning() const { return periodic_hanning_; }
// The padding type used for FFT.
FftPaddingType GetFftPaddingType() const { return fft_padding_type_; }
// Setter APIs.
void SetSampleRateHz(int sample_rate_hz) { sample_rate_hz_ = sample_rate_hz; }
void SetNumChannels(int num_channels) { num_channels_ = num_channels; }
void SetInputScale(float input_scale) { input_scale_ = input_scale; }
void SetPreEmphasisFactor(float pre_emphasis_factor) {
pre_emphasis_factor_ = pre_emphasis_factor;
}
// The FFT length must be even for real FFT optimization. The FFT bins will be
// derived from the FFT length as FFT length / 2 + 1.
void SetFftLength(int fft_length) {
fft_length_ = fft_length;
fft_bins_ = fft_length / 2 + 1;
}
void SetFrameLength(int frame_length) { frame_length_ = frame_length; }
void SetHopLength(int hop_length) { hop_length_ = hop_length; }
void SetNumMelBins(int num_mel_bins) { num_mel_bins_ = num_mel_bins; }
void SetMelLowHz(float mel_low_hz) { mel_low_hz_ = mel_low_hz; }
void SetMelHighHz(float mel_high_hz) { mel_high_hz_ = mel_high_hz; }
void SetMelFloor(float mel_floor) { mel_floor_ = mel_floor; }
void SetNormalizeMel(bool normalize_mel) { normalize_mel_ = normalize_mel; }
void SetAddFloorToMelBeforeLog(bool add_floor_to_mel_before_log) {
add_floor_to_mel_before_log_ = add_floor_to_mel_before_log;
}
void SetSemicausalPadding(bool semicausal_padding) {
semicausal_padding_ = semicausal_padding;
}
void SetNonZeroHanning(bool non_zero_hanning) {
non_zero_hanning_ = non_zero_hanning;
}
void SetPeriodicHanning(bool periodic_hanning) {
periodic_hanning_ = periodic_hanning;
}
void SetFftPaddingType(FftPaddingType fft_padding_type) {
fft_padding_type_ = fft_padding_type;
}
// The Mel Spectrogram means used for Universal Speech Model (USM) during
// preprocessing.
static constexpr std::array<float, 128> kUsmMelMean{
6.398797734146062, 6.5292966718485665, 6.636971307272159,
6.73283598251503, 6.83729192594687, 6.955722303271236,
7.102944890730766, 7.114182036087843, 7.1506544101153,
7.174958993259514, 7.1890256978077804, 7.196835788986042,
7.211737590554171, 7.365040287042535, 7.350661707754529,
7.34752702412618, 7.370936184320344, 7.552167274579683,
7.4736985912567455, 7.461733145619613, 7.655010083032587,
7.537023586741711, 7.59332033698754, 7.678828995158089,
7.573545549481997, 7.721706263812856, 7.548489195294597,
7.647480899467908, 7.546350507038094, 7.552359044394656,
7.60142267532906, 7.510803537242497, 7.547512749381739,
7.5734628575808145, 7.516065818981327, 7.544310572169082,
7.556128732606547, 7.578428971230521, 7.565946473157099,
7.565821431053628, 7.582146705201401, 7.5917054493764775,
7.59647680034444, 7.612909043144701, 7.642191074647679,
7.682020208604412, 7.669657702288002, 7.636762908696176,
7.645613169792156, 7.687786852309006, 7.733375349074729,
7.705414197270183, 7.773851002316419, 7.767855696186511,
7.804625030416079, 7.8095583241565505, 7.845300151068656,
7.832030482713495, 7.876477438621265, 7.886595835981996,
7.907747879286325, 7.926010325946424, 7.927971987569718,
7.94765994925662, 7.9609369675109205, 7.977485334083968,
7.995276449058029, 8.020093867153456, 8.026893789702653,
8.036394113138993, 8.072079269745391, 8.072009510709744,
8.15832987882215, 8.169035932109242, 8.201262910500471,
8.203176911295596, 8.237251381186532, 8.265968214462914,
8.278791003594298, 8.279921657260331, 8.303751782080207,
8.323985266369666, 8.358499418073363, 8.368121771923692,
8.392162333974197, 8.40529917133684, 8.421934604788884,
8.43307981480797, 8.416437732709245, 8.380481381138022,
8.313028108945332, 8.172698101608145, 7.987087868524417,
7.775018865353218, 7.587469885918491, 7.485680948258058,
7.425561455270659, 7.426161453764725, 7.500171657170674,
7.473711809407939, 7.497915553109761, 7.555291079941853,
7.5404297094497155, 7.554637855844384, 7.5536294881940025,
7.597411437015373, 7.620857310821611, 7.622024042245356,
7.643684482318661, 7.651806604022742, 7.647768200868812,
7.619968160658521, 7.663675433728041, 7.770133777809638,
7.775737195054957, 7.756637821283381, 7.7958903182806445,
7.824714343764584, 7.8699194044250325, 7.857690367947652,
7.854133456399421, 7.83057312917979, 7.780062155284722,
7.687571300835443, 7.626255596158039, 7.475138444832542,
7.31241576045514, 7.162930372619685,
};
// The Mel Spectrogram standard deviations constants used for Universal Speech
// Model (USM) during preprocessing.
static constexpr std::array<float, 128> kUsmMelStdDev{
1.6785894541269812, 1.6687138672328043, 1.6906522689607268,
1.7375192957945016, 1.7755335232132188, 1.7945350399969586,
1.8160038735261768, 1.8455822079478754, 1.854889301328728,
1.8544058257314018, 1.8531530795826658, 1.8568193392072,
1.8568580559801775, 1.8403822120311448, 1.8311156303932052,
1.8381223837390877, 1.8582757939740133, 1.8751353033960765,
1.8940031697532662, 1.9045566324594227, 1.9114104933328382,
1.9234409916967738, 1.932244372950416, 1.9354540832886058,
1.9196173248258872, 1.8884371698304272, 1.8666212011400265,
1.851852265212217, 1.8466309429379515, 1.8370433682382064,
1.8312948374209728, 1.8233918348681029, 1.8162900339615862,
1.813554336166136, 1.7988012203002604, 1.7783664628243725,
1.762995373099593, 1.754638830337111, 1.7562192553046327,
1.7570134298011308, 1.748103676233597, 1.7420266564237143,
1.7433799765791382, 1.7405273444710188, 1.7681605535143332,
1.7928765468247894, 1.7832784911754684, 1.7556019331853459,
1.734978397119943, 1.7251193027145706, 1.711577677561937,
1.7077475454470532, 1.702793505675667, 1.7087228728780646,
1.7055479598955696, 1.7048659481569446, 1.7136985315687527,
1.7003759527643025, 1.7038510617369829, 1.712407460050622,
1.7195395708962748, 1.715985369102956, 1.7047382463157097,
1.6858892841332958, 1.6803980138770978, 1.6883086163746897,
1.678822586089551, 1.6704169259147215, 1.6824154866833487,
1.7002006169486261, 1.7095077608591729, 1.7127719919531275,
1.7007540237588394, 1.7007030789334565, 1.7006801726721705,
1.7084333739135957, 1.7080081837410785, 1.7088852843730529,
1.7058124003569382, 1.7104967128913229, 1.7017088898161998,
1.6946290530635235, 1.6886895951157692, 1.6913609136330663,
1.6802034976166595, 1.6778644057956866, 1.6844856225324205,
1.6919889285341483, 1.6918548241011255, 1.6771215766236411,
1.6753742459089904, 1.6732896439517075, 1.665104739745144,
1.682512689327978, 1.7001049276791989, 1.71496232533367,
1.751371703351037, 1.7589949482516734, 1.7274831977280356,
1.7428303906628124, 1.7427952258580872, 1.7072930970436015,
1.72696991469254, 1.7128335116767701, 1.7266508365456639,
1.699287147275948, 1.6860698274507981, 1.6862991003373358,
1.683393071329867, 1.687619365543026, 1.7100825041856975,
1.7407356256589301, 1.7218710733945026, 1.6776658140019411,
1.6864518015922916, 1.7273244787326472, 1.6992470398169233,
1.6800806970795965, 1.6579370965601807, 1.6647055065206582,
1.65766768806214, 1.6294301234765352, 1.5918612004781831,
1.5335441292387613, 1.3949765253217616, 1.2628815962896491,
1.1053653031914006, 0.9263256925938697,
};
private:
explicit AudioPreprocessorConfig(
// Audio decoding parameters.
int sample_rate_hz, int num_channels,
// FFT parameters.
int frame_length, int hop_length, int fft_length, float input_scale,
float pre_emphasis_factor,
// Mel spectrogram parameters.
int num_mel_bins, float mel_low_hz, float mel_high_hz, float mel_floor,
bool normalize_mel, bool add_floor_to_mel_before_log,
bool semicausal_padding, bool non_zero_hanning, bool periodic_hanning,
FftPaddingType fft_padding_type)
: sample_rate_hz_(sample_rate_hz),
num_channels_(num_channels),
fft_length_(fft_length),
fft_bins_(fft_length / 2 + 1),
frame_length_(frame_length),
hop_length_(hop_length),
num_mel_bins_(num_mel_bins),
mel_low_hz_(mel_low_hz),
mel_high_hz_(mel_high_hz),
mel_floor_(mel_floor),
input_scale_(input_scale),
pre_emphasis_factor_(pre_emphasis_factor),
normalize_mel_(normalize_mel),
add_floor_to_mel_before_log_(add_floor_to_mel_before_log),
semicausal_padding_(semicausal_padding),
non_zero_hanning_(non_zero_hanning),
periodic_hanning_(periodic_hanning),
fft_padding_type_(fft_padding_type) {}
int sample_rate_hz_;
int num_channels_;
int fft_length_;
int fft_bins_;
int frame_length_;
int hop_length_;
int num_mel_bins_;
float mel_low_hz_;
float mel_high_hz_;
float mel_floor_;
float input_scale_;
float pre_emphasis_factor_;
bool normalize_mel_;
bool add_floor_to_mel_before_log_;
bool semicausal_padding_;
bool non_zero_hanning_;
bool periodic_hanning_;
FftPaddingType fft_padding_type_;
};
// Interface for audio preprocessing.
class AudioPreprocessor {
public:
virtual ~AudioPreprocessor() = default;
// Preprocesses the undecoded audio bytes and returns the preprocessed audio.
virtual absl::StatusOr<InputAudio> Preprocess(
const InputAudio& input_audio) = 0;
// Reset the audio preprocessor to the initial state.
virtual void Reset() = 0;
};
std::ostream& operator<<(
std::ostream& os,
const AudioPreprocessorConfig::FftPaddingType& padding_type);
} // namespace litert::lm
#endif // THIRD_PARTY_ODML_LITERT_LM_RUNTIME_COMPONENTS_PREPROCESSOR_AUDIO_PREPROCESSOR_H_
|