Bill Cox | dfe6b37 | 2011-07-15 13:38:26 -0400 | [diff] [blame] | 1 | /* Sonic library |
| 2 | Copyright 2010, 2011 |
| 3 | Bill Cox |
| 4 | This file is part of the Sonic Library. |
| 5 | |
Bill Cox | 60eeb06 | 2015-02-27 10:17:45 -0800 | [diff] [blame] | 6 | This file is licensed under the Apache 2.0 license. |
Bill Cox | e720065 | 2011-07-16 12:09:05 -0400 | [diff] [blame] | 7 | */ |
Bill Cox | dfe6b37 | 2011-07-15 13:38:26 -0400 | [diff] [blame] | 8 | |
| 9 | package sonic; |
| 10 | |
| 11 | public class Sonic { |
| 12 | |
| 13 | private static final int SONIC_MIN_PITCH = 65; |
| 14 | private static final int SONIC_MAX_PITCH = 400; |
| 15 | /* This is used to down-sample some inputs to improve speed */ |
| 16 | private static final int SONIC_AMDF_FREQ = 4000; |
| 17 | |
| 18 | private short inputBuffer[]; |
| 19 | private short outputBuffer[]; |
| 20 | private short pitchBuffer[]; |
| 21 | private short downSampleBuffer[]; |
| 22 | private float speed; |
| 23 | private float volume; |
| 24 | private float pitch; |
| 25 | private float rate; |
| 26 | private int oldRatePosition; |
| 27 | private int newRatePosition; |
| 28 | private boolean useChordPitch; |
| 29 | private int quality; |
| 30 | private int numChannels; |
| 31 | private int inputBufferSize; |
| 32 | private int pitchBufferSize; |
| 33 | private int outputBufferSize; |
| 34 | private int numInputSamples; |
| 35 | private int numOutputSamples; |
| 36 | private int numPitchSamples; |
| 37 | private int minPeriod; |
| 38 | private int maxPeriod; |
| 39 | private int maxRequired; |
| 40 | private int remainingInputToCopy; |
| 41 | private int sampleRate; |
| 42 | private int prevPeriod; |
| 43 | private int prevMinDiff; |
| 44 | |
| 45 | // Resize the array. |
| 46 | private short[] resize( |
| 47 | short[] oldArray, |
| 48 | int newLength) |
| 49 | { |
David R. Souto | 3225150 | 2013-09-10 13:33:34 +0200 | [diff] [blame] | 50 | newLength *= numChannels; |
Bill Cox | dfe6b37 | 2011-07-15 13:38:26 -0400 | [diff] [blame] | 51 | short[] newArray = new short[newLength]; |
| 52 | int length = oldArray.length <= newLength? oldArray.length : newLength; |
| 53 | |
David R. Souto | 3225150 | 2013-09-10 13:33:34 +0200 | [diff] [blame] | 54 | |
Bill Cox | dfe6b37 | 2011-07-15 13:38:26 -0400 | [diff] [blame] | 55 | for(int x = 0; x < length; x++) { |
| 56 | newArray[x] = oldArray[x]; |
| 57 | } |
| 58 | return newArray; |
| 59 | } |
| 60 | |
| 61 | // Move samples from one array to another. May move samples down within an array, but not up. |
| 62 | private void move( |
| 63 | short dest[], |
| 64 | int destPos, |
| 65 | short source[], |
| 66 | int sourcePos, |
| 67 | int numSamples) |
| 68 | { |
| 69 | for(int xSample = 0; xSample < numSamples*numChannels; xSample++) { |
| 70 | dest[destPos*numChannels + xSample] = source[sourcePos*numChannels + xSample]; |
| 71 | } |
| 72 | } |
| 73 | |
| 74 | // Scale the samples by the factor. |
| 75 | private void scaleSamples( |
| 76 | short samples[], |
| 77 | int position, |
| 78 | int numSamples, |
| 79 | float volume) |
| 80 | { |
| 81 | int fixedPointVolume = (int)(volume*4096.0f); |
| 82 | int start = position*numChannels; |
| 83 | int stop = start + numSamples*numChannels; |
| 84 | |
| 85 | for(int xSample = start; xSample < stop; xSample++) { |
| 86 | int value = (samples[xSample]*fixedPointVolume) >> 12; |
| 87 | if(value > 32767) { |
| 88 | value = 32767; |
| 89 | } else if(value < -32767) { |
| 90 | value = -32767; |
| 91 | } |
| 92 | samples[xSample] = (short)value; |
| 93 | } |
| 94 | } |
| 95 | |
| 96 | // Get the speed of the stream. |
| 97 | public float getSpeed() |
| 98 | { |
| 99 | return speed; |
| 100 | } |
| 101 | |
| 102 | // Set the speed of the stream. |
| 103 | public void setSpeed( |
| 104 | float speed) |
| 105 | { |
| 106 | this.speed = speed; |
| 107 | } |
| 108 | |
| 109 | // Get the pitch of the stream. |
| 110 | public float getPitch() |
| 111 | { |
| 112 | return pitch; |
| 113 | } |
| 114 | |
| 115 | // Set the pitch of the stream. |
| 116 | public void setPitch( |
| 117 | float pitch) |
| 118 | { |
| 119 | this.pitch = pitch; |
| 120 | } |
| 121 | |
| 122 | // Get the rate of the stream. |
| 123 | public float getRate() |
| 124 | { |
| 125 | return rate; |
| 126 | } |
| 127 | |
| 128 | // Set the playback rate of the stream. This scales pitch and speed at the same time. |
| 129 | public void setRate( |
| 130 | float rate) |
| 131 | { |
| 132 | this.rate = rate; |
| 133 | this.oldRatePosition = 0; |
| 134 | this.newRatePosition = 0; |
| 135 | } |
| 136 | |
| 137 | // Get the vocal chord pitch setting. |
| 138 | public boolean getChordPitch() |
| 139 | { |
| 140 | return useChordPitch; |
| 141 | } |
| 142 | |
| 143 | // Set the vocal chord mode for pitch computation. Default is off. |
| 144 | public void setChordPitch( |
| 145 | boolean useChordPitch) |
| 146 | { |
| 147 | this.useChordPitch = useChordPitch; |
| 148 | } |
| 149 | |
| 150 | // Get the quality setting. |
| 151 | public int getQuality() |
| 152 | { |
| 153 | return quality; |
| 154 | } |
| 155 | |
| 156 | // Set the "quality". Default 0 is virtually as good as 1, but very much faster. |
| 157 | public void setQuality( |
| 158 | int quality) |
| 159 | { |
| 160 | this.quality = quality; |
| 161 | } |
| 162 | |
| 163 | // Get the scaling factor of the stream. |
| 164 | public float getVolume() |
| 165 | { |
| 166 | return volume; |
| 167 | } |
| 168 | |
| 169 | // Set the scaling factor of the stream. |
| 170 | public void setVolume( |
| 171 | float volume) |
| 172 | { |
| 173 | this.volume = volume; |
| 174 | } |
| 175 | |
| 176 | // Allocate stream buffers. |
| 177 | private void allocateStreamBuffers( |
| 178 | int sampleRate, |
| 179 | int numChannels) |
| 180 | { |
| 181 | minPeriod = sampleRate/SONIC_MAX_PITCH; |
| 182 | maxPeriod = sampleRate/SONIC_MIN_PITCH; |
| 183 | maxRequired = 2*maxPeriod; |
| 184 | inputBufferSize = maxRequired; |
| 185 | inputBuffer = new short[maxRequired*numChannels]; |
| 186 | outputBufferSize = maxRequired; |
| 187 | outputBuffer = new short[maxRequired*numChannels]; |
| 188 | pitchBufferSize = maxRequired; |
| 189 | pitchBuffer = new short[maxRequired*numChannels]; |
| 190 | downSampleBuffer = new short[maxRequired]; |
| 191 | this.sampleRate = sampleRate; |
| 192 | this.numChannels = numChannels; |
| 193 | oldRatePosition = 0; |
| 194 | newRatePosition = 0; |
| 195 | prevPeriod = 0; |
| 196 | } |
| 197 | |
| 198 | // Create a sonic stream. |
| 199 | public Sonic( |
| 200 | int sampleRate, |
| 201 | int numChannels) |
| 202 | { |
| 203 | allocateStreamBuffers(sampleRate, numChannels); |
| 204 | speed = 1.0f; |
| 205 | pitch = 1.0f; |
| 206 | volume = 1.0f; |
| 207 | rate = 1.0f; |
| 208 | oldRatePosition = 0; |
| 209 | newRatePosition = 0; |
| 210 | useChordPitch = false; |
| 211 | quality = 0; |
| 212 | } |
| 213 | |
| 214 | // Get the sample rate of the stream. |
| 215 | public int getSampleRate() |
| 216 | { |
| 217 | return sampleRate; |
| 218 | } |
| 219 | |
| 220 | // Set the sample rate of the stream. This will cause samples buffered in the stream to be lost. |
| 221 | public void setSampleRate( |
| 222 | int sampleRate) |
| 223 | { |
| 224 | allocateStreamBuffers(sampleRate, numChannels); |
| 225 | } |
| 226 | |
| 227 | // Get the number of channels. |
| 228 | public int getNumChannels() |
| 229 | { |
| 230 | return numChannels; |
| 231 | } |
| 232 | |
| 233 | // Set the num channels of the stream. This will cause samples buffered in the stream to be lost. |
| 234 | public void setNumChannels( |
| 235 | int numChannels) |
| 236 | { |
| 237 | allocateStreamBuffers(sampleRate, numChannels); |
| 238 | } |
| 239 | |
| 240 | // Enlarge the output buffer if needed. |
| 241 | private void enlargeOutputBufferIfNeeded( |
| 242 | int numSamples) |
| 243 | { |
| 244 | if(numOutputSamples + numSamples > outputBufferSize) { |
| 245 | outputBufferSize += (outputBufferSize >> 1) + numSamples; |
| 246 | outputBuffer = resize(outputBuffer, outputBufferSize); |
| 247 | } |
| 248 | } |
| 249 | |
| 250 | // Enlarge the input buffer if needed. |
| 251 | private void enlargeInputBufferIfNeeded( |
| 252 | int numSamples) |
| 253 | { |
| 254 | if(numInputSamples + numSamples > inputBufferSize) { |
| 255 | inputBufferSize += (inputBufferSize >> 1) + numSamples; |
| 256 | inputBuffer = resize(inputBuffer, inputBufferSize); |
| 257 | } |
| 258 | } |
| 259 | |
| 260 | // Add the input samples to the input buffer. |
| 261 | private void addFloatSamplesToInputBuffer( |
| 262 | float samples[], |
| 263 | int numSamples) |
| 264 | { |
| 265 | if(numSamples == 0) { |
| 266 | return; |
| 267 | } |
| 268 | enlargeInputBufferIfNeeded(numSamples); |
| 269 | int xBuffer = numInputSamples*numChannels; |
| 270 | for(int xSample = 0; xSample < numSamples*numChannels; xSample++) { |
| 271 | inputBuffer[xBuffer++] = (short)(samples[xSample]*32767.0f); |
| 272 | } |
| 273 | numInputSamples += numSamples; |
| 274 | } |
| 275 | |
| 276 | // Add the input samples to the input buffer. |
| 277 | private void addShortSamplesToInputBuffer( |
| 278 | short samples[], |
| 279 | int numSamples) |
| 280 | { |
| 281 | if(numSamples == 0) { |
| 282 | return; |
| 283 | } |
| 284 | enlargeInputBufferIfNeeded(numSamples); |
| 285 | move(inputBuffer, numInputSamples, samples, 0, numSamples); |
| 286 | numInputSamples += numSamples; |
| 287 | } |
| 288 | |
| 289 | // Add the input samples to the input buffer. |
| 290 | private void addUnsignedByteSamplesToInputBuffer( |
| 291 | byte samples[], |
| 292 | int numSamples) |
| 293 | { |
| 294 | short sample; |
| 295 | |
| 296 | enlargeInputBufferIfNeeded(numSamples); |
| 297 | int xBuffer = numInputSamples*numChannels; |
| 298 | for(int xSample = 0; xSample < numSamples*numChannels; xSample++) { |
| 299 | sample = (short)((samples[xSample] & 0xff) - 128); // Convert from unsigned to signed |
| 300 | inputBuffer[xBuffer++] = (short) (sample << 8); |
| 301 | } |
| 302 | numInputSamples += numSamples; |
| 303 | } |
| 304 | |
Bill Cox | 2e48c22 | 2011-07-16 12:00:46 -0400 | [diff] [blame] | 305 | // Add the input samples to the input buffer. They must be 16-bit little-endian encoded in a byte array. |
| 306 | private void addBytesToInputBuffer( |
| 307 | byte inBuffer[], |
| 308 | int numBytes) |
| 309 | { |
| 310 | int numSamples = numBytes/(2*numChannels); |
| 311 | short sample; |
| 312 | |
| 313 | enlargeInputBufferIfNeeded(numSamples); |
| 314 | int xBuffer = numInputSamples*numChannels; |
| 315 | for(int xByte = 0; xByte + 1 < numBytes; xByte += 2) { |
| 316 | sample = (short)((inBuffer[xByte] & 0xff) | (inBuffer[xByte + 1] << 8)); |
| 317 | inputBuffer[xBuffer++] = sample; |
| 318 | } |
| 319 | numInputSamples += numSamples; |
| 320 | } |
| 321 | |
Bill Cox | dfe6b37 | 2011-07-15 13:38:26 -0400 | [diff] [blame] | 322 | // Remove input samples that we have already processed. |
| 323 | private void removeInputSamples( |
| 324 | int position) |
| 325 | { |
| 326 | int remainingSamples = numInputSamples - position; |
| 327 | |
| 328 | move(inputBuffer, 0, inputBuffer, position, remainingSamples); |
| 329 | numInputSamples = remainingSamples; |
| 330 | } |
| 331 | |
| 332 | // Just copy from the array to the output buffer |
| 333 | private void copyToOutput( |
| 334 | short samples[], |
| 335 | int position, |
| 336 | int numSamples) |
| 337 | { |
| 338 | enlargeOutputBufferIfNeeded(numSamples); |
| 339 | move(outputBuffer, numOutputSamples, samples, position, numSamples); |
| 340 | numOutputSamples += numSamples; |
| 341 | } |
| 342 | |
| 343 | // Just copy from the input buffer to the output buffer. Return num samples copied. |
| 344 | private int copyInputToOutput( |
| 345 | int position) |
| 346 | { |
| 347 | int numSamples = remainingInputToCopy; |
| 348 | |
| 349 | if(numSamples > maxRequired) { |
| 350 | numSamples = maxRequired; |
| 351 | } |
| 352 | copyToOutput(inputBuffer, position, numSamples); |
| 353 | remainingInputToCopy -= numSamples; |
| 354 | return numSamples; |
| 355 | } |
| 356 | |
| 357 | // Read data out of the stream. Sometimes no data will be available, and zero |
| 358 | // is returned, which is not an error condition. |
| 359 | public int readFloatFromStream( |
| 360 | float samples[], |
| 361 | int maxSamples) |
| 362 | { |
| 363 | int numSamples = numOutputSamples; |
| 364 | int remainingSamples = 0; |
| 365 | |
| 366 | if(numSamples == 0) { |
| 367 | return 0; |
| 368 | } |
| 369 | if(numSamples > maxSamples) { |
| 370 | remainingSamples = numSamples - maxSamples; |
| 371 | numSamples = maxSamples; |
| 372 | } |
| 373 | for(int xSample = 0; xSample < numSamples*numChannels; xSample++) { |
| 374 | samples[xSample++] = (outputBuffer[xSample])/32767.0f; |
| 375 | } |
| 376 | move(outputBuffer, 0, outputBuffer, numSamples, remainingSamples); |
| 377 | numOutputSamples = remainingSamples; |
| 378 | return numSamples; |
| 379 | } |
| 380 | |
| 381 | // Read short data out of the stream. Sometimes no data will be available, and zero |
| 382 | // is returned, which is not an error condition. |
| 383 | public int readShortFromStream( |
| 384 | short samples[], |
| 385 | int maxSamples) |
| 386 | { |
| 387 | int numSamples = numOutputSamples; |
| 388 | int remainingSamples = 0; |
| 389 | |
| 390 | if(numSamples == 0) { |
| 391 | return 0; |
| 392 | } |
| 393 | if(numSamples > maxSamples) { |
| 394 | remainingSamples = numSamples - maxSamples; |
| 395 | numSamples = maxSamples; |
| 396 | } |
| 397 | move(samples, 0, outputBuffer, 0, numSamples); |
| 398 | move(outputBuffer, 0, outputBuffer, numSamples, remainingSamples); |
| 399 | numOutputSamples = remainingSamples; |
| 400 | return numSamples; |
| 401 | } |
| 402 | |
| 403 | // Read unsigned byte data out of the stream. Sometimes no data will be available, and zero |
| 404 | // is returned, which is not an error condition. |
| 405 | public int readUnsignedByteFromStream( |
| 406 | byte samples[], |
| 407 | int maxSamples) |
| 408 | { |
| 409 | int numSamples = numOutputSamples; |
| 410 | int remainingSamples = 0; |
| 411 | |
| 412 | if(numSamples == 0) { |
| 413 | return 0; |
| 414 | } |
| 415 | if(numSamples > maxSamples) { |
| 416 | remainingSamples = numSamples - maxSamples; |
| 417 | numSamples = maxSamples; |
| 418 | } |
| 419 | for(int xSample = 0; xSample < numSamples*numChannels; xSample++) { |
| 420 | samples[xSample] = (byte)((outputBuffer[xSample] >> 8) + 128); |
| 421 | } |
| 422 | move(outputBuffer, 0, outputBuffer, numSamples, remainingSamples); |
| 423 | numOutputSamples = remainingSamples; |
| 424 | return numSamples; |
| 425 | } |
| 426 | |
Bill Cox | 2e48c22 | 2011-07-16 12:00:46 -0400 | [diff] [blame] | 427 | // Read unsigned byte data out of the stream. Sometimes no data will be available, and zero |
| 428 | // is returned, which is not an error condition. |
| 429 | public int readBytesFromStream( |
| 430 | byte outBuffer[], |
| 431 | int maxBytes) |
| 432 | { |
| 433 | int maxSamples = maxBytes/(2*numChannels); |
| 434 | int numSamples = numOutputSamples; |
| 435 | int remainingSamples = 0; |
| 436 | |
| 437 | if(numSamples == 0 || maxSamples == 0) { |
| 438 | return 0; |
| 439 | } |
| 440 | if(numSamples > maxSamples) { |
| 441 | remainingSamples = numSamples - maxSamples; |
| 442 | numSamples = maxSamples; |
| 443 | } |
| 444 | for(int xSample = 0; xSample < numSamples*numChannels; xSample++) { |
| 445 | short sample = outputBuffer[xSample]; |
| 446 | outBuffer[xSample << 1] = (byte)(sample & 0xff); |
| 447 | outBuffer[(xSample << 1) + 1] = (byte)(sample >> 8); |
| 448 | } |
| 449 | move(outputBuffer, 0, outputBuffer, numSamples, remainingSamples); |
| 450 | numOutputSamples = remainingSamples; |
| 451 | return 2*numSamples*numChannels; |
| 452 | } |
| 453 | |
Bill Cox | dfe6b37 | 2011-07-15 13:38:26 -0400 | [diff] [blame] | 454 | // Force the sonic stream to generate output using whatever data it currently |
| 455 | // has. No extra delay will be added to the output, but flushing in the middle of |
| 456 | // words could introduce distortion. |
| 457 | public void flushStream() |
| 458 | { |
| 459 | int remainingSamples = numInputSamples; |
| 460 | float s = speed/pitch; |
| 461 | float r = rate*pitch; |
| 462 | int expectedOutputSamples = numOutputSamples + (int)((remainingSamples/s + numPitchSamples)/r + 0.5f); |
| 463 | |
| 464 | // Add enough silence to flush both input and pitch buffers. |
| 465 | enlargeInputBufferIfNeeded(remainingSamples + 2*maxRequired); |
| 466 | for(int xSample = 0; xSample < 2*maxRequired*numChannels; xSample++) { |
| 467 | inputBuffer[remainingSamples*numChannels + xSample] = 0; |
| 468 | } |
| 469 | numInputSamples += 2*maxRequired; |
| 470 | writeShortToStream(null, 0); |
| 471 | // Throw away any extra samples we generated due to the silence we added. |
| 472 | if(numOutputSamples > expectedOutputSamples) { |
| 473 | numOutputSamples = expectedOutputSamples; |
| 474 | } |
| 475 | // Empty input and pitch buffers. |
| 476 | numInputSamples = 0; |
| 477 | remainingInputToCopy = 0; |
| 478 | numPitchSamples = 0; |
| 479 | } |
| 480 | |
| 481 | // Return the number of samples in the output buffer |
| 482 | public int samplesAvailable() |
| 483 | { |
| 484 | return numOutputSamples; |
| 485 | } |
| 486 | |
| 487 | // If skip is greater than one, average skip samples together and write them to |
| 488 | // the down-sample buffer. If numChannels is greater than one, mix the channels |
| 489 | // together as we down sample. |
| 490 | private void downSampleInput( |
| 491 | short samples[], |
| 492 | int position, |
| 493 | int skip) |
| 494 | { |
| 495 | int numSamples = maxRequired/skip; |
| 496 | int samplesPerValue = numChannels*skip; |
| 497 | int value; |
| 498 | |
| 499 | position *= numChannels; |
| 500 | for(int i = 0; i < numSamples; i++) { |
| 501 | value = 0; |
| 502 | for(int j = 0; j < samplesPerValue; j++) { |
| 503 | value += samples[position + i*samplesPerValue + j]; |
| 504 | } |
| 505 | value /= samplesPerValue; |
| 506 | downSampleBuffer[i] = (short)value; |
| 507 | } |
| 508 | } |
| 509 | |
| 510 | // Find the best frequency match in the range, and given a sample skip multiple. |
| 511 | // For now, just find the pitch of the first channel. Note that retMinDiff and |
| 512 | // retMaxDiff are Int objects, which the caller will need to create with new. |
| 513 | private int findPitchPeriodInRange( |
| 514 | short samples[], |
| 515 | int position, |
| 516 | int minPeriod, |
| 517 | int maxPeriod, |
| 518 | Integer retMinDiff, |
| 519 | Integer retMaxDiff) |
| 520 | { |
| 521 | int bestPeriod = 0, worstPeriod = 255; |
| 522 | int minDiff = 1, maxDiff = 0; |
| 523 | |
| 524 | position *= numChannels; |
| 525 | for(int period = minPeriod; period <= maxPeriod; period++) { |
| 526 | int diff = 0; |
| 527 | for(int i = 0; i < period; i++) { |
| 528 | short sVal = samples[position + i]; |
| 529 | short pVal = samples[position + period + i]; |
| 530 | diff += sVal >= pVal? sVal - pVal : pVal - sVal; |
| 531 | } |
| 532 | /* Note that the highest number of samples we add into diff will be less |
| 533 | than 256, since we skip samples. Thus, diff is a 24 bit number, and |
| 534 | we can safely multiply by numSamples without overflow */ |
| 535 | if(diff*bestPeriod < minDiff*period) { |
| 536 | minDiff = diff; |
| 537 | bestPeriod = period; |
| 538 | } |
| 539 | if(diff*worstPeriod > maxDiff*period) { |
| 540 | maxDiff = diff; |
| 541 | worstPeriod = period; |
| 542 | } |
| 543 | } |
| 544 | retMinDiff = minDiff/bestPeriod; |
| 545 | retMaxDiff = maxDiff/worstPeriod; |
| 546 | return bestPeriod; |
| 547 | } |
| 548 | |
| 549 | // At abrupt ends of voiced words, we can have pitch periods that are better |
| 550 | // approximated by the previous pitch period estimate. Try to detect this case. |
| 551 | private boolean prevPeriodBetter( |
| 552 | int period, |
| 553 | int minDiff, |
| 554 | int maxDiff, |
| 555 | boolean preferNewPeriod) |
| 556 | { |
| 557 | if(minDiff == 0 || prevPeriod == 0) { |
| 558 | return false; |
| 559 | } |
| 560 | if(preferNewPeriod) { |
| 561 | if(maxDiff > minDiff*3) { |
| 562 | // Got a reasonable match this period |
| 563 | return false; |
| 564 | } |
| 565 | if(minDiff*2 <= prevMinDiff*3) { |
| 566 | // Mismatch is not that much greater this period |
| 567 | return false; |
| 568 | } |
| 569 | } else { |
| 570 | if(minDiff <= prevMinDiff) { |
| 571 | return false; |
| 572 | } |
| 573 | } |
| 574 | return true; |
| 575 | } |
| 576 | |
| 577 | // Find the pitch period. This is a critical step, and we may have to try |
| 578 | // multiple ways to get a good answer. This version uses AMDF. To improve |
| 579 | // speed, we down sample by an integer factor get in the 11KHz range, and then |
| 580 | // do it again with a narrower frequency range without down sampling |
| 581 | private int findPitchPeriod( |
| 582 | short samples[], |
| 583 | int position, |
| 584 | boolean preferNewPeriod) |
| 585 | { |
| 586 | Integer minDiff = new Integer(0); |
| 587 | Integer maxDiff = new Integer(0); |
| 588 | int period, retPeriod; |
| 589 | int skip = 1; |
| 590 | |
| 591 | if(sampleRate > SONIC_AMDF_FREQ && quality == 0) { |
| 592 | skip = sampleRate/SONIC_AMDF_FREQ; |
| 593 | } |
| 594 | if(numChannels == 1 && skip == 1) { |
| 595 | period = findPitchPeriodInRange(samples, position, minPeriod, maxPeriod, minDiff, maxDiff); |
| 596 | } else { |
| 597 | downSampleInput(samples, position, skip); |
| 598 | period = findPitchPeriodInRange(downSampleBuffer, 0, minPeriod/skip, |
| 599 | maxPeriod/skip, minDiff, maxDiff); |
| 600 | if(skip != 1) { |
| 601 | period *= skip; |
| 602 | int minP = period - (skip << 2); |
| 603 | int maxP = period + (skip << 2); |
| 604 | if(minP < minPeriod) { |
| 605 | minP = minPeriod; |
| 606 | } |
| 607 | if(maxP > maxPeriod) { |
| 608 | maxP = maxPeriod; |
| 609 | } |
| 610 | if(numChannels == 1) { |
| 611 | period = findPitchPeriodInRange(samples, position, minP, maxP, minDiff, maxDiff); |
| 612 | } else { |
| 613 | downSampleInput(samples, position, 1); |
| 614 | period = findPitchPeriodInRange(downSampleBuffer, 0, minP, maxP, minDiff, maxDiff); |
| 615 | } |
| 616 | } |
| 617 | } |
| 618 | if(prevPeriodBetter(period, minDiff, maxDiff, preferNewPeriod)) { |
| 619 | retPeriod = prevPeriod; |
| 620 | } else { |
| 621 | retPeriod = period; |
| 622 | } |
| 623 | prevMinDiff = minDiff; |
| 624 | prevPeriod = period; |
| 625 | return retPeriod; |
| 626 | } |
| 627 | |
| 628 | // Overlap two sound segments, ramp the volume of one down, while ramping the |
| 629 | // other one from zero up, and add them, storing the result at the output. |
| 630 | private void overlapAdd( |
| 631 | int numSamples, |
| 632 | int numChannels, |
| 633 | short out[], |
| 634 | int outPos, |
| 635 | short rampDown[], |
| 636 | int rampDownPos, |
| 637 | short rampUp[], |
| 638 | int rampUpPos) |
| 639 | { |
| 640 | for(int i = 0; i < numChannels; i++) { |
| 641 | int o = outPos*numChannels + i; |
| 642 | int u = rampUpPos*numChannels + i; |
| 643 | int d = rampDownPos*numChannels + i; |
| 644 | for(int t = 0; t < numSamples; t++) { |
| 645 | out[o] = (short)((rampDown[d]*(numSamples - t) + rampUp[u]*t)/numSamples); |
| 646 | o += numChannels; |
| 647 | d += numChannels; |
| 648 | u += numChannels; |
| 649 | } |
| 650 | } |
| 651 | } |
| 652 | |
| 653 | // Overlap two sound segments, ramp the volume of one down, while ramping the |
| 654 | // other one from zero up, and add them, storing the result at the output. |
| 655 | private void overlapAddWithSeparation( |
| 656 | int numSamples, |
| 657 | int numChannels, |
| 658 | int separation, |
| 659 | short out[], |
| 660 | int outPos, |
| 661 | short rampDown[], |
| 662 | int rampDownPos, |
| 663 | short rampUp[], |
| 664 | int rampUpPos) |
| 665 | { |
| 666 | for(int i = 0; i < numChannels; i++) { |
| 667 | int o = outPos*numChannels + i; |
| 668 | int u = rampUpPos*numChannels + i; |
| 669 | int d = rampDownPos*numChannels + i; |
| 670 | for(int t = 0; t < numSamples + separation; t++) { |
| 671 | if(t < separation) { |
| 672 | out[o] = (short)(rampDown[d]*(numSamples - t)/numSamples); |
| 673 | d += numChannels; |
| 674 | } else if(t < numSamples) { |
| 675 | out[o] = (short)((rampDown[d]*(numSamples - t) + rampUp[u]*(t - separation))/numSamples); |
| 676 | d += numChannels; |
| 677 | u += numChannels; |
| 678 | } else { |
| 679 | out[o] = (short)(rampUp[u]*(t - separation)/numSamples); |
| 680 | u += numChannels; |
| 681 | } |
| 682 | o += numChannels; |
| 683 | } |
| 684 | } |
| 685 | } |
| 686 | |
| 687 | // Just move the new samples in the output buffer to the pitch buffer |
| 688 | private void moveNewSamplesToPitchBuffer( |
| 689 | int originalNumOutputSamples) |
| 690 | { |
| 691 | int numSamples = numOutputSamples - originalNumOutputSamples; |
| 692 | |
| 693 | if(numPitchSamples + numSamples > pitchBufferSize) { |
| 694 | pitchBufferSize += (pitchBufferSize >> 1) + numSamples; |
| 695 | pitchBuffer = resize(pitchBuffer, pitchBufferSize); |
| 696 | } |
| 697 | move(pitchBuffer, numPitchSamples, outputBuffer, originalNumOutputSamples, numSamples); |
| 698 | numOutputSamples = originalNumOutputSamples; |
| 699 | numPitchSamples += numSamples; |
| 700 | } |
| 701 | |
| 702 | // Remove processed samples from the pitch buffer. |
| 703 | private void removePitchSamples( |
| 704 | int numSamples) |
| 705 | { |
| 706 | if(numSamples == 0) { |
| 707 | return; |
| 708 | } |
| 709 | move(pitchBuffer, 0, pitchBuffer, numSamples, numPitchSamples - numSamples); |
| 710 | numPitchSamples -= numSamples; |
| 711 | } |
| 712 | |
| 713 | // Change the pitch. The latency this introduces could be reduced by looking at |
| 714 | // past samples to determine pitch, rather than future. |
| 715 | private void adjustPitch( |
| 716 | int originalNumOutputSamples) |
| 717 | { |
| 718 | int period, newPeriod, separation; |
| 719 | int position = 0; |
| 720 | |
| 721 | if(numOutputSamples == originalNumOutputSamples) { |
| 722 | return; |
| 723 | } |
| 724 | moveNewSamplesToPitchBuffer(originalNumOutputSamples); |
| 725 | while(numPitchSamples - position >= maxRequired) { |
| 726 | period = findPitchPeriod(pitchBuffer, position, false); |
| 727 | newPeriod = (int)(period/pitch); |
| 728 | enlargeOutputBufferIfNeeded(newPeriod); |
| 729 | if(pitch >= 1.0f) { |
| 730 | overlapAdd(newPeriod, numChannels, outputBuffer, numOutputSamples, pitchBuffer, |
| 731 | position, pitchBuffer, position + period - newPeriod); |
| 732 | } else { |
| 733 | separation = newPeriod - period; |
| 734 | overlapAddWithSeparation(period, numChannels, separation, outputBuffer, numOutputSamples, |
| 735 | pitchBuffer, position, pitchBuffer, position); |
| 736 | } |
| 737 | numOutputSamples += newPeriod; |
| 738 | position += period; |
| 739 | } |
| 740 | removePitchSamples(position); |
| 741 | } |
| 742 | |
| 743 | // Interpolate the new output sample. |
| 744 | private short interpolate( |
| 745 | short in[], |
| 746 | int inPos, |
| 747 | int oldSampleRate, |
| 748 | int newSampleRate) |
| 749 | { |
| 750 | short left = in[inPos*numChannels]; |
| 751 | short right = in[inPos*numChannels + numChannels]; |
| 752 | int position = newRatePosition*oldSampleRate; |
| 753 | int leftPosition = oldRatePosition*newSampleRate; |
| 754 | int rightPosition = (oldRatePosition + 1)*newSampleRate; |
| 755 | int ratio = rightPosition - position; |
| 756 | int width = rightPosition - leftPosition; |
| 757 | |
| 758 | return (short)((ratio*left + (width - ratio)*right)/width); |
| 759 | } |
| 760 | |
| 761 | // Change the rate. |
| 762 | private void adjustRate( |
| 763 | float rate, |
| 764 | int originalNumOutputSamples) |
| 765 | { |
| 766 | int newSampleRate = (int)(sampleRate/rate); |
| 767 | int oldSampleRate = sampleRate; |
| 768 | int position; |
| 769 | |
| 770 | // Set these values to help with the integer math |
| 771 | while(newSampleRate > (1 << 14) || oldSampleRate > (1 << 14)) { |
| 772 | newSampleRate >>= 1; |
| 773 | oldSampleRate >>= 1; |
| 774 | } |
| 775 | if(numOutputSamples == originalNumOutputSamples) { |
| 776 | return; |
| 777 | } |
| 778 | moveNewSamplesToPitchBuffer(originalNumOutputSamples); |
| 779 | // Leave at least one pitch sample in the buffer |
| 780 | for(position = 0; position < numPitchSamples - 1; position++) { |
| 781 | while((oldRatePosition + 1)*newSampleRate > newRatePosition*oldSampleRate) { |
| 782 | enlargeOutputBufferIfNeeded(1); |
| 783 | for(int i = 0; i < numChannels; i++) { |
| 784 | outputBuffer[numOutputSamples*numChannels + i] = interpolate(pitchBuffer, position + i, |
| 785 | oldSampleRate, newSampleRate); |
| 786 | } |
| 787 | newRatePosition++; |
| 788 | numOutputSamples++; |
| 789 | } |
| 790 | oldRatePosition++; |
| 791 | if(oldRatePosition == oldSampleRate) { |
| 792 | oldRatePosition = 0; |
| 793 | if(newRatePosition != newSampleRate) { |
| 794 | System.out.printf("Assertion failed: newRatePosition != newSampleRate\n"); |
| 795 | assert false; |
| 796 | } |
| 797 | newRatePosition = 0; |
| 798 | } |
| 799 | } |
| 800 | removePitchSamples(position); |
| 801 | } |
| 802 | |
| 803 | |
| 804 | // Skip over a pitch period, and copy period/speed samples to the output |
| 805 | private int skipPitchPeriod( |
| 806 | short samples[], |
| 807 | int position, |
| 808 | float speed, |
| 809 | int period) |
| 810 | { |
| 811 | int newSamples; |
| 812 | |
| 813 | if(speed >= 2.0f) { |
| 814 | newSamples = (int)(period/(speed - 1.0f)); |
| 815 | } else { |
| 816 | newSamples = period; |
| 817 | remainingInputToCopy = (int)(period*(2.0f - speed)/(speed - 1.0f)); |
| 818 | } |
| 819 | enlargeOutputBufferIfNeeded(newSamples); |
| 820 | overlapAdd(newSamples, numChannels, outputBuffer, numOutputSamples, samples, position, |
| 821 | samples, position + period); |
| 822 | numOutputSamples += newSamples; |
| 823 | return newSamples; |
| 824 | } |
| 825 | |
| 826 | // Insert a pitch period, and determine how much input to copy directly. |
| 827 | private int insertPitchPeriod( |
| 828 | short samples[], |
| 829 | int position, |
| 830 | float speed, |
| 831 | int period) |
| 832 | { |
| 833 | int newSamples; |
| 834 | |
| 835 | if(speed < 0.5f) { |
| 836 | newSamples = (int)(period*speed/(1.0f - speed)); |
| 837 | } else { |
| 838 | newSamples = period; |
| 839 | remainingInputToCopy = (int)(period*(2.0f*speed - 1.0f)/(1.0f - speed)); |
| 840 | } |
| 841 | enlargeOutputBufferIfNeeded(period + newSamples); |
| 842 | move(outputBuffer, numOutputSamples, samples, position, period); |
| 843 | overlapAdd(newSamples, numChannels, outputBuffer, numOutputSamples + period, samples, |
| 844 | position + period, samples, position); |
| 845 | numOutputSamples += period + newSamples; |
| 846 | return newSamples; |
| 847 | } |
| 848 | |
| 849 | // Resample as many pitch periods as we have buffered on the input. Return 0 if |
| 850 | // we fail to resize an input or output buffer. Also scale the output by the volume. |
| 851 | private void changeSpeed( |
| 852 | float speed) |
| 853 | { |
| 854 | int numSamples = numInputSamples; |
| 855 | int position = 0, period, newSamples; |
| 856 | |
| 857 | if(numInputSamples < maxRequired) { |
| 858 | return; |
| 859 | } |
| 860 | do { |
| 861 | if(remainingInputToCopy > 0) { |
| 862 | newSamples = copyInputToOutput(position); |
| 863 | position += newSamples; |
| 864 | } else { |
| 865 | period = findPitchPeriod(inputBuffer, position, true); |
| 866 | if(speed > 1.0) { |
| 867 | newSamples = skipPitchPeriod(inputBuffer, position, speed, period); |
| 868 | position += period + newSamples; |
| 869 | } else { |
| 870 | newSamples = insertPitchPeriod(inputBuffer, position, speed, period); |
| 871 | position += newSamples; |
| 872 | } |
| 873 | } |
| 874 | } while(position + maxRequired <= numSamples); |
| 875 | removeInputSamples(position); |
| 876 | } |
| 877 | |
| 878 | // Resample as many pitch periods as we have buffered on the input. Scale the output by the volume. |
| 879 | private void processStreamInput() |
| 880 | { |
| 881 | int originalNumOutputSamples = numOutputSamples; |
| 882 | float s = speed/pitch; |
| 883 | float r = rate; |
| 884 | |
| 885 | if(!useChordPitch) { |
| 886 | r *= pitch; |
| 887 | } |
| 888 | if(s > 1.00001 || s < 0.99999) { |
Bill Cox | 2e48c22 | 2011-07-16 12:00:46 -0400 | [diff] [blame] | 889 | changeSpeed(s); |
Bill Cox | dfe6b37 | 2011-07-15 13:38:26 -0400 | [diff] [blame] | 890 | } else { |
| 891 | copyToOutput(inputBuffer, 0, numInputSamples); |
| 892 | numInputSamples = 0; |
| 893 | } |
| 894 | if(useChordPitch) { |
| 895 | if(pitch != 1.0f) { |
| 896 | adjustPitch(originalNumOutputSamples); |
| 897 | } |
Bill Cox | 2e48c22 | 2011-07-16 12:00:46 -0400 | [diff] [blame] | 898 | } else if(r != 1.0f) { |
| 899 | adjustRate(r, originalNumOutputSamples); |
Bill Cox | dfe6b37 | 2011-07-15 13:38:26 -0400 | [diff] [blame] | 900 | } |
| 901 | if(volume != 1.0f) { |
| 902 | // Adjust output volume. |
| 903 | scaleSamples(outputBuffer, originalNumOutputSamples, numOutputSamples - originalNumOutputSamples, |
| 904 | volume); |
| 905 | } |
| 906 | } |
| 907 | |
| 908 | // Write floating point data to the input buffer and process it. |
| 909 | public void writeFloatToStream( |
| 910 | float samples[], |
| 911 | int numSamples) |
| 912 | { |
| 913 | addFloatSamplesToInputBuffer(samples, numSamples); |
| 914 | processStreamInput(); |
| 915 | } |
| 916 | |
| 917 | // Write the data to the input stream, and process it. |
| 918 | public void writeShortToStream( |
| 919 | short samples[], |
| 920 | int numSamples) |
| 921 | { |
| 922 | addShortSamplesToInputBuffer(samples, numSamples); |
| 923 | processStreamInput(); |
| 924 | } |
| 925 | |
| 926 | // Simple wrapper around sonicWriteFloatToStream that does the unsigned byte to short |
| 927 | // conversion for you. |
| 928 | public void writeUnsignedByteToStream( |
| 929 | byte samples[], |
| 930 | int numSamples) |
| 931 | { |
| 932 | addUnsignedByteSamplesToInputBuffer(samples, numSamples); |
| 933 | processStreamInput(); |
| 934 | } |
| 935 | |
Bill Cox | 2e48c22 | 2011-07-16 12:00:46 -0400 | [diff] [blame] | 936 | // Simple wrapper around sonicWriteBytesToStream that does the byte to 16-bit LE conversion. |
| 937 | public void writeBytesToStream( |
| 938 | byte inBuffer[], |
| 939 | int numBytes) |
| 940 | { |
| 941 | addBytesToInputBuffer(inBuffer, numBytes); |
| 942 | processStreamInput(); |
| 943 | } |
| 944 | |
Bill Cox | dfe6b37 | 2011-07-15 13:38:26 -0400 | [diff] [blame] | 945 | // This is a non-stream oriented interface to just change the speed of a sound sample |
| 946 | public static int changeFloatSpeed( |
| 947 | float samples[], |
| 948 | int numSamples, |
| 949 | float speed, |
| 950 | float pitch, |
| 951 | float rate, |
| 952 | float volume, |
| 953 | boolean useChordPitch, |
| 954 | int sampleRate, |
| 955 | int numChannels) |
| 956 | { |
| 957 | Sonic stream = new Sonic(sampleRate, numChannels); |
| 958 | |
| 959 | stream.setSpeed(speed); |
| 960 | stream.setPitch(pitch); |
| 961 | stream.setRate(rate); |
| 962 | stream.setVolume(volume); |
| 963 | stream.setChordPitch(useChordPitch); |
| 964 | stream.writeFloatToStream(samples, numSamples); |
| 965 | stream.flushStream(); |
| 966 | numSamples = stream.samplesAvailable(); |
| 967 | stream.readFloatFromStream(samples, numSamples); |
| 968 | return numSamples; |
| 969 | } |
| 970 | |
| 971 | /* This is a non-stream oriented interface to just change the speed of a sound sample */ |
| 972 | public int sonicChangeShortSpeed( |
| 973 | short samples[], |
| 974 | int numSamples, |
| 975 | float speed, |
| 976 | float pitch, |
| 977 | float rate, |
| 978 | float volume, |
| 979 | boolean useChordPitch, |
| 980 | int sampleRate, |
| 981 | int numChannels) |
| 982 | { |
| 983 | Sonic stream = new Sonic(sampleRate, numChannels); |
| 984 | |
| 985 | stream.setSpeed(speed); |
| 986 | stream.setPitch(pitch); |
| 987 | stream.setRate(rate); |
| 988 | stream.setVolume(volume); |
| 989 | stream.setChordPitch(useChordPitch); |
| 990 | stream.writeShortToStream(samples, numSamples); |
| 991 | stream.flushStream(); |
| 992 | numSamples = stream.samplesAvailable(); |
| 993 | stream.readShortFromStream(samples, numSamples); |
| 994 | return numSamples; |
| 995 | } |
Bill Cox | e720065 | 2011-07-16 12:09:05 -0400 | [diff] [blame] | 996 | } |