[315] | 1 | package org.apollo.meldex;
|
---|
| 2 |
|
---|
| 3 | import java.io.IOException;
|
---|
| 4 | import java.util.ArrayList;
|
---|
| 5 |
|
---|
| 6 | public class Transcriber
|
---|
| 7 | {
|
---|
| 8 | // The values of these constants are taken directly from Rodger's code
|
---|
| 9 | final double RMS_WINDOW = 10.0;
|
---|
| 10 | final double RMS_STEP = (RMS_WINDOW / 2);
|
---|
| 11 |
|
---|
| 12 | // The sample rate (frequency) of the loaded sample
|
---|
| 13 | int sampleRate;
|
---|
| 14 |
|
---|
| 15 | // Create instances of our classes
|
---|
| 16 | PitchTracker pitchTracker = new PitchTracker();
|
---|
| 17 | NoteRounder noteRounder = new NoteRounder();
|
---|
| 18 |
|
---|
| 19 |
|
---|
| 20 | /**
|
---|
| 21 | *
|
---|
| 22 | * @param sample
|
---|
| 23 | *
|
---|
| 24 | * @return
|
---|
| 25 | * The wave sample converted into a RogTrack structure if successful.
|
---|
| 26 | * Null if unsuccessful.
|
---|
| 27 | *
|
---|
| 28 | * @throws IOException
|
---|
| 29 | * If conversion to standardized form fails
|
---|
| 30 | */
|
---|
| 31 | public RogTrack transcribeSample(WavSample sample) throws IOException
|
---|
| 32 | {
|
---|
| 33 | // This shouldn't have to be fixed...
|
---|
| 34 | sampleRate = 22050;
|
---|
| 35 | // sampleRate = (int)SampledAudioManager.PLAYBACK_SAMPLE_RATE;
|
---|
| 36 |
|
---|
| 37 | // These are values that should end up being parameters
|
---|
| 38 | int tempoBPM = 120;
|
---|
| 39 | int minRestLength = 4, minNoteLength = 2;
|
---|
| 40 | boolean addRestLeftovers = true;
|
---|
| 41 |
|
---|
| 42 | // Check that the sample is valid
|
---|
| 43 | if (sample == null || sample.getFormat() == null || sample.getRawAudio() == null) {
|
---|
| 44 | return null;
|
---|
| 45 | }
|
---|
| 46 |
|
---|
| 47 | // Get the standardised unsigned 8-bit mono data
|
---|
| 48 | //byte[] stdData = sample.getStandardisedData(true); // THis is bad of memory - was getting lots heap errors
|
---|
| 49 | // AudioFormat standardizedFormat = new AudioFormat(
|
---|
| 50 | // //sample.getFormat().getSampleRate(),
|
---|
| 51 | // sampleRate,
|
---|
| 52 | // //sample.getFormat().getSampleRate(),
|
---|
| 53 | // 8, // 8-bit
|
---|
| 54 | // 1, // mono
|
---|
| 55 | // false, // unsigned
|
---|
| 56 | // true); // Big endian
|
---|
| 57 | //
|
---|
| 58 | // byte[] stdData = AudioIO.convertAudioBytes(
|
---|
| 59 | // sample.getRawAudio(),
|
---|
| 60 | // sample.getFormat(),
|
---|
| 61 | // standardizedFormat);
|
---|
| 62 |
|
---|
| 63 | byte[] stdData = MeldexConversion.toStandardizedFormat(
|
---|
| 64 | sample.getRawAudio(), sample.getFormat(), sampleRate);
|
---|
| 65 |
|
---|
| 66 | if (stdData == null) {
|
---|
| 67 | return null;
|
---|
| 68 | }
|
---|
| 69 |
|
---|
| 70 | // Create a new RogTrack to store the result of the transcription
|
---|
| 71 | RogTrack track = new RogTrack();
|
---|
| 72 |
|
---|
| 73 | // We don't know the time signature
|
---|
| 74 | track.addTimeSignature(0, 0);
|
---|
| 75 |
|
---|
| 76 | // Calculate the Root Mean Squared data for the loaded sample
|
---|
| 77 | float[] rmsData = calculateRMS(stdData, stdData.length);
|
---|
| 78 |
|
---|
| 79 | // Check that we have some data to process
|
---|
| 80 | if (rmsData.length <= 0) {
|
---|
| 81 | return null;
|
---|
| 82 | }
|
---|
| 83 |
|
---|
| 84 |
|
---|
| 85 | // int rmsStep = 111; // !! PURE-ROG!!
|
---|
| 86 | int rmsStep = (sampleRate / 200);
|
---|
| 87 | //int rmsStep = (sampleRate / 1000);
|
---|
| 88 |
|
---|
| 89 |
|
---|
| 90 | // Calculate the number of RMS steps per sixteenth note
|
---|
| 91 | int rmsStepsPer16th = (int) ((60F / (float) (tempoBPM * 4F)) / (RMS_STEP / 1000F));
|
---|
| 92 |
|
---|
| 93 | // Calculate the sum of the squared RMS data
|
---|
| 94 | double sumRMSDataSquared = 0;
|
---|
| 95 | for (int rmsPos = 0; rmsPos < rmsData.length; rmsPos++) {
|
---|
| 96 | sumRMSDataSquared += (rmsData[rmsPos] * rmsData[rmsPos]);
|
---|
| 97 | }
|
---|
| 98 |
|
---|
| 99 | // Calculate the upper and lower thresholds (asymmetrical??)
|
---|
| 100 | double threshold = Math.sqrt(sumRMSDataSquared / rmsData.length);
|
---|
| 101 | float upperThreshold = (float) (threshold * 0.55);
|
---|
| 102 | float lowerThreshold = (float) (threshold * 0.35);
|
---|
| 103 |
|
---|
| 104 | // Now we find all of the notes in the sample...
|
---|
| 105 | int lastCents = -1, lastStart = 0, lastEnd = 0;
|
---|
| 106 | int restLen, noteLen;
|
---|
| 107 | for (int rmsPos = 0; rmsPos < rmsData.length; rmsPos++) {
|
---|
| 108 |
|
---|
| 109 | // Have we found the start of a note??
|
---|
| 110 | if (rmsData[rmsPos] >= upperThreshold) {
|
---|
| 111 |
|
---|
| 112 | // Yes, so find the end of the note
|
---|
| 113 | int noteStart = rmsPos;
|
---|
| 114 | while (rmsPos < rmsData.length) {
|
---|
| 115 | if (rmsData[rmsPos] <= lowerThreshold) {
|
---|
| 116 | break;
|
---|
| 117 | }
|
---|
| 118 | rmsPos++;
|
---|
| 119 | }
|
---|
| 120 |
|
---|
| 121 | int noteEnd = rmsPos;
|
---|
| 122 | // If it is longer than 1/3 of the minimum note length, process it
|
---|
| 123 | if ((noteEnd - noteStart) >= ((minNoteLength * rmsStepsPer16th) / 3)) {
|
---|
| 124 |
|
---|
| 125 | // Create a new array for the interesting data
|
---|
| 126 | int noteLength = (noteEnd - noteStart) * rmsStep;
|
---|
| 127 | byte[] noteData = new byte[noteLength];
|
---|
| 128 | for (int notePos = 0; notePos < noteLength; notePos++) {
|
---|
| 129 | noteData[notePos] = stdData[(noteStart * rmsStep) + notePos];
|
---|
| 130 | }
|
---|
| 131 |
|
---|
| 132 | // Process the note
|
---|
| 133 | int cents = processNote(noteData, noteLength);
|
---|
| 134 | if (cents > 1000) {
|
---|
| 135 |
|
---|
| 136 | // If it is the first note then initialise the NoteRounder
|
---|
| 137 | if (lastCents < 0) {
|
---|
| 138 | int noteRound = (noteRounder.roundCents(cents) - cents);
|
---|
| 139 | noteRounder.setInitialValue(noteRound);
|
---|
| 140 | }
|
---|
| 141 | // Otherwise we calculate the note and rest values
|
---|
| 142 | else {
|
---|
| 143 | int finalNote = noteRounder.roundNote(lastCents, true);
|
---|
| 144 |
|
---|
| 145 | // Calculate the rest length
|
---|
| 146 | int rmsRestLeft = 0;
|
---|
| 147 | if (rmsStepsPer16th <= 0) {
|
---|
| 148 | restLen = 0;
|
---|
| 149 | }
|
---|
| 150 | else {
|
---|
| 151 | int rmsRestLen = (noteStart - lastEnd);
|
---|
| 152 | int mult = (minRestLength * rmsStepsPer16th);
|
---|
| 153 | int roundRest = noteRounder.round(rmsRestLen, mult);
|
---|
| 154 | restLen = (roundRest / rmsStepsPer16th);
|
---|
| 155 | if (addRestLeftovers == true) {
|
---|
| 156 | rmsRestLeft = rmsRestLen - (restLen * rmsStepsPer16th);
|
---|
| 157 | }
|
---|
| 158 | }
|
---|
| 159 |
|
---|
| 160 | // Calculate the note length
|
---|
| 161 | if (rmsStepsPer16th <= 0) {
|
---|
| 162 | noteLen = 0;
|
---|
| 163 | }
|
---|
| 164 | else {
|
---|
| 165 | int rmsNoteLen = (lastEnd - lastStart);
|
---|
| 166 | int mult = (minNoteLength * rmsStepsPer16th);
|
---|
| 167 | if (addRestLeftovers == true && rmsRestLeft > 0) {
|
---|
| 168 | rmsNoteLen += rmsRestLeft;
|
---|
| 169 | }
|
---|
| 170 | int roundNote = noteRounder.round(rmsNoteLen, mult);
|
---|
| 171 | noteLen = (roundNote / rmsStepsPer16th);
|
---|
| 172 | }
|
---|
| 173 |
|
---|
| 174 | // Add the notes and rests to the track
|
---|
| 175 | track.addNote(noteLen, (finalNote / 100), 0);
|
---|
| 176 | track.addRest(restLen);
|
---|
| 177 | }
|
---|
| 178 |
|
---|
| 179 | lastCents = cents;
|
---|
| 180 | lastStart = noteStart;
|
---|
| 181 | lastEnd = noteEnd;
|
---|
| 182 | }
|
---|
| 183 | }
|
---|
| 184 | }
|
---|
| 185 | }
|
---|
| 186 |
|
---|
| 187 | // Deal with the last note (if there is one)
|
---|
| 188 | if (lastCents > 0) {
|
---|
| 189 | int finalNote = noteRounder.roundNote(lastCents, true);
|
---|
| 190 |
|
---|
| 191 | // Calculate the note length
|
---|
| 192 | if (rmsStepsPer16th <= 0) {
|
---|
| 193 | noteLen = 0;
|
---|
| 194 | }
|
---|
| 195 | else {
|
---|
| 196 | int rmsNoteLen = (lastEnd - lastStart);
|
---|
| 197 | int mult = (minNoteLength * rmsStepsPer16th);
|
---|
| 198 | int roundNote = noteRounder.round(rmsNoteLen, mult);
|
---|
| 199 | noteLen = (roundNote / rmsStepsPer16th);
|
---|
| 200 |
|
---|
| 201 | // Add the last note to the track
|
---|
| 202 | track.addNote(noteLen, (finalNote / 100), 0);
|
---|
| 203 | }
|
---|
| 204 | }
|
---|
| 205 |
|
---|
| 206 | // Calculate the key signature
|
---|
| 207 | int key = track.calculateBestKey();
|
---|
| 208 |
|
---|
| 209 | // Add a key signature event to the track
|
---|
| 210 | track.addKeySignature(key, 0);
|
---|
| 211 |
|
---|
| 212 | // Set the note names for the new key signature
|
---|
| 213 | track.setNoteNames(key);
|
---|
| 214 |
|
---|
| 215 | // Return the Track with the transcribed data
|
---|
| 216 | return track;
|
---|
| 217 | }
|
---|
| 218 |
|
---|
| 219 |
|
---|
| 220 | private float[] calculateRMS(byte[] stdData, int stdLength)
|
---|
| 221 | {
|
---|
| 222 | int accMS = 0;
|
---|
| 223 | int accMSPrev = 0;
|
---|
| 224 | int frameStart = 0;
|
---|
| 225 |
|
---|
| 226 | // int rmsStep = 111; // !! PURE ROG!!
|
---|
| 227 | int rmsStep = (22050 / 200); // !! SEMI-PURE ROG !!
|
---|
| 228 |
|
---|
| 229 | // Allocate memory for the RMS data values
|
---|
| 230 | int rmsLength = (int) (stdLength / (float) rmsStep);
|
---|
| 231 | float[] rmsData = new float[rmsLength];
|
---|
| 232 |
|
---|
| 233 | for (int stdPos = 0, rmsPos = 0; stdPos < stdLength; stdPos++) {
|
---|
| 234 | // Calculate the accumulated Mean Squared value
|
---|
| 235 | int value = (stdData[stdPos] & 255) - 127; // !! SHOULD 127 BE 128?? !!
|
---|
| 236 | accMS += (value * value);
|
---|
| 237 |
|
---|
| 238 | // If we have finished a frame calculate a value for the output
|
---|
| 239 | if ((stdPos - frameStart) >= rmsStep) {
|
---|
| 240 | double result = Math.sqrt((double)(accMS+accMSPrev) / (double)(rmsStep*2));
|
---|
| 241 | frameStart = stdPos;
|
---|
| 242 | accMSPrev = accMS;
|
---|
| 243 | accMS = 0;
|
---|
| 244 | rmsData[rmsPos] = (float) result;
|
---|
| 245 | rmsPos++;
|
---|
| 246 | }
|
---|
| 247 | }
|
---|
| 248 |
|
---|
| 249 | return rmsData;
|
---|
| 250 | }
|
---|
| 251 |
|
---|
| 252 |
|
---|
| 253 | private int processNote(byte[] data, int length)
|
---|
| 254 | {
|
---|
| 255 | // Pitch track the note
|
---|
[1102] | 256 | ArrayList<PitchValue> pitchData = pitchTracker.process(data, length, sampleRate);
|
---|
[315] | 257 |
|
---|
| 258 | // Average the pitch data
|
---|
| 259 | int pitchLength = averagePitchData(pitchData);
|
---|
| 260 |
|
---|
| 261 | // Calculate the histogram of the pitch data
|
---|
| 262 | return calculateHistogram(pitchData, pitchLength);
|
---|
| 263 | }
|
---|
| 264 |
|
---|
| 265 |
|
---|
[1102] | 266 | private int averagePitchData(ArrayList<PitchValue> pitchData)
|
---|
[315] | 267 | {
|
---|
| 268 | // Loop through the pitch values...
|
---|
| 269 | int i = 0, k = 0;
|
---|
| 270 | while (i < (pitchData.size() - 1)) {
|
---|
[1102] | 271 | int startPos = pitchData.get(i).position;
|
---|
| 272 | double period = pitchData.get(i).period;
|
---|
[315] | 273 | double averagePeriod = period;
|
---|
[1102] | 274 | double runningPeriod = pitchData.get(i+1).position - startPos;
|
---|
[315] | 275 | double numPeriods = runningPeriod / period;
|
---|
| 276 |
|
---|
| 277 | // Start from the next pitch
|
---|
| 278 | int j = 1;
|
---|
| 279 | while ((i+j) < (pitchData.size() - 1)) {
|
---|
| 280 | // Get the next pitch estimate
|
---|
[1102] | 281 | period = pitchData.get(i+j).period;
|
---|
| 282 | int position = pitchData.get(i+j).position;
|
---|
[315] | 283 |
|
---|
| 284 | // Make sure that this period is covered by the average so far
|
---|
| 285 | if (period > (averagePeriod * 1.1) || period < (averagePeriod * 0.909)) {
|
---|
| 286 | break;
|
---|
| 287 | }
|
---|
| 288 |
|
---|
| 289 | // Stop if we have covered more than 20 msec
|
---|
[1102] | 290 | int nextPos = pitchData.get(i+j+1).position;
|
---|
[315] | 291 | // if ((nextPos - startPos) >= 445) { // !! PURE-ROG !!
|
---|
| 292 | if ((nextPos - startPos) >= (sampleRate * 0.02)) {
|
---|
| 293 | break;
|
---|
| 294 | }
|
---|
| 295 |
|
---|
| 296 | // Add to this period
|
---|
| 297 | runningPeriod += (nextPos - position);
|
---|
| 298 | numPeriods += ((nextPos - position) / period);
|
---|
| 299 | averagePeriod = runningPeriod / numPeriods;
|
---|
| 300 | j++;
|
---|
| 301 | }
|
---|
| 302 |
|
---|
| 303 | // Write over the original pitch value with the averaged pitch value
|
---|
| 304 | pitchData.set(k, new PitchValue(averagePeriod, startPos));
|
---|
| 305 | k++;
|
---|
| 306 |
|
---|
| 307 | // Increment i
|
---|
| 308 | i += j;
|
---|
| 309 | }
|
---|
| 310 |
|
---|
| 311 | // Return the number of averaged pitch values
|
---|
| 312 | return k;
|
---|
| 313 | }
|
---|
| 314 |
|
---|
| 315 |
|
---|
[1102] | 316 | private int calculateHistogram(ArrayList<PitchValue> data, int length)
|
---|
[315] | 317 | {
|
---|
| 318 | // This probably shouldn't be a constant
|
---|
| 319 | final int histLength = 960;
|
---|
| 320 |
|
---|
| 321 | // Allocate memory for the histogram data and initialise it to zero
|
---|
| 322 | int[] histData = new int[histLength];
|
---|
| 323 | for (int i = 0; i < histLength; i++) {
|
---|
| 324 | histData[i] = 0;
|
---|
| 325 | }
|
---|
| 326 |
|
---|
| 327 | // Calculate the histogram data
|
---|
| 328 | for (int i = 0; i < (length - 1); i++) {
|
---|
[1102] | 329 | double period = data.get(i).period;
|
---|
| 330 | double position = data.get(i).position;
|
---|
| 331 | double periodLength = data.get(i+1).position - position;
|
---|
[315] | 332 |
|
---|
| 333 | // We only need to do this if the pitch does not equal zero
|
---|
| 334 | if (period > 0) {
|
---|
| 335 | // Do some weird stuff...
|
---|
| 336 | double logValue = (Math.log(period / sampleRate)) / (Math.log(10));
|
---|
| 337 | double cents = -(logValue / 0.000250858) - 3637.622659;
|
---|
| 338 | int histPos = (int) (Math.floor((cents - 3600.0) / 5.0));
|
---|
| 339 | if (histPos >= 0 && histPos < histLength) {
|
---|
| 340 | histData[histPos] += periodLength;
|
---|
| 341 | }
|
---|
| 342 | }
|
---|
| 343 | }
|
---|
| 344 |
|
---|
| 345 | // Select the cents value using the histogram data
|
---|
| 346 | return selectCents(histData, histLength);
|
---|
| 347 | }
|
---|
| 348 |
|
---|
| 349 |
|
---|
| 350 | private int selectCents(int[] histData, int histLength)
|
---|
| 351 | {
|
---|
| 352 | // Find the maximum average histogram value
|
---|
| 353 | int maxHistValue = 0, maxHistPos = 0;
|
---|
| 354 | for (int i = 0; i < (histLength - 20); i++) {
|
---|
| 355 | // Sum the next 20 histogram values (1 semitone)
|
---|
| 356 | int histValue = 0;
|
---|
| 357 | for (int j = 0; j < 20; j++) {
|
---|
| 358 | histValue += histData[i+j];
|
---|
| 359 | }
|
---|
| 360 |
|
---|
| 361 | // Maintain the maximum histogram value
|
---|
| 362 | if (histValue > maxHistValue) {
|
---|
| 363 | maxHistValue = histValue;
|
---|
| 364 | maxHistPos = i;
|
---|
| 365 | }
|
---|
| 366 | }
|
---|
| 367 |
|
---|
| 368 | // Find the average cents value of the maximum histogram
|
---|
| 369 | int sumCents = 0;
|
---|
| 370 | for (int j = 0; j < 20; j++) {
|
---|
| 371 | sumCents += (histData[maxHistPos+j] * (((maxHistPos+j) * 5) + 3602));
|
---|
| 372 | }
|
---|
| 373 |
|
---|
| 374 | // Calculate sum of (cents * time) divided by total time
|
---|
| 375 | if (maxHistValue > 0) {
|
---|
| 376 | return (sumCents / maxHistValue);
|
---|
| 377 | }
|
---|
| 378 | else {
|
---|
| 379 | return 0;
|
---|
| 380 | }
|
---|
| 381 | }
|
---|
| 382 | }
|
---|