Spaces:
Sleeping
Sleeping
File size: 19,065 Bytes
460123d e154249 ddefe81 3aeb1fc 460123d d97e211 460123d e38cada 460123d ddefe81 f262643 eec853e a1d4ef1 eec853e 709f33f eec853e b305e6a ddefe81 bc2fc31 ddefe81 e1cef83 ddefe81 699e931 ddefe81 e1cef83 e708b94 e1cef83 ddefe81 e1cef83 ddefe81 dbd0dd8 bc2fc31 ddefe81 bc2fc31 ddefe81 3aeb1fc 7860d14 853ad6e 7860d14 3bd2a42 7860d14 3bd2a42 7860d14 3bd2a42 7860d14 3bd2a42 7860d14 3bd2a42 7860d14 3bd2a42 7860d14 3bd2a42 7860d14 6233f8b 7860d14 612bfe6 7860d14 612bfe6 7860d14 e851785 7860d14 e851785 7860d14 432c704 e851785 7860d14 e851785 7860d14 e851785 7860d14 e851785 7860d14 e851785 7860d14 e851785 7860d14 9b477b6 7860d14 b1c27b9 a73ce4b 7860d14 b1c27b9 7860d14 b1c27b9 7860d14 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 |
import cv2
import random
import copy
from pyannote.core import Annotation, Segment
import numpy as np
import torch
import torchaudio
import pandas as pd
import datetime as dt
def colors(n):
'''
Creates a list size n of distinctive colors
'''
if n == 0:
return []
ret = []
h = int(random.random() * 180)
step = 180 / n
for i in range(n):
h += step
h = int(h) % 180
hsv = np.uint8([[[h,200,200]]])
bgr = cv2.cvtColor(hsv,cv2.COLOR_HSV2BGR)
ret.append((bgr[0][0][0].item()/255,bgr[0][0][1].item()/255,bgr[0][0][2].item()/255))
return ret
def colorsCSS(n):
'''
Creates a list size n of distinctive colors based on CSS formatting
'''
if n == 0:
return []
ret = []
h = int(random.random() * 180)
step = 180 / n
for i in range(n):
h += step
h = int(h) % 180
hsv = np.uint8([[[h,200,200]]])
bgr = cv2.cvtColor(hsv,cv2.COLOR_HSV2BGR)
b = f'{bgr[0][0][0].item():02x}'
g = f'{bgr[0][0][1].item():02x}'
r = f'{bgr[0][0][2].item():02x}'
ret.append('#'+b+g+r)
return ret
def extendSpeakers(mySpeakerList, fileLabel = 'NONE', maximumSecondDifference = 1, minimumSecondDuration = 0):
'''
Assumes mySpeakerList is already split into Speaker/Audience
'''
mySpeakerAnnotations = Annotation(uri=fileLabel)
newSpeakerList = [[],[]]
for i, speaker in enumerate(mySpeakerList):
speaker.sort()
lastEnd = -1
tempSection = None
for section in speaker:
if lastEnd == -1:
tempSection = copy.deepcopy(section)
lastEnd = section[0] + section[1]
else:
if section[0] - lastEnd <= maximumSecondDifference:
tempSection = (tempSection[0],max(section[0] + section[1] - tempSection[0],tempSection[1]))
lastEnd = tempSection[0] + tempSection[1]
else:
if tempSection[1] >= minimumSecondDuration:
newSpeakerList[i].append(tempSection)
mySpeakerAnnotations[Segment(tempSection[0],lastEnd)] = i
tempSection = copy.deepcopy(section)
lastEnd = section[0] + section[1]
if tempSection is not None:
# Add the last section back in
if tempSection[1] >= minimumSecondDuration:
newSpeakerList[i].append(tempSection)
mySpeakerAnnotations[Segment(tempSection[0],lastEnd)] = i
return newSpeakerList,mySpeakerAnnotations
def twoClassExtendAnnotation(myAnnotation,maximumSecondDifference = 1, minimumSecondDuration = 0):
lecturerID = None
lecturerLen = 0
# Identify lecturer
for speakerName in myAnnotation.labels():
tempLen = len(myAnnotation.label_support(speakerName))
if tempLen > lecturerLen:
lecturerLen = tempLen
lecturerID = speakerName
tempSpeakerList = [[],[]]
# Recreate speakerList as [[lecturer labels],[audience labels]]
for speakerName in myAnnotation.labels():
if speakerName != lecturerID:
for segmentItem in myAnnotation.label_support(speakerName):
tempSpeakerList[1].append((segmentItem.start,segmentItem.duration))
else:
for segmentItem in myAnnotation.label_support(speakerName):
tempSpeakerList[0].append((segmentItem.start,segmentItem.duration))
newList, newAnnotation = extendSpeakers(tempSpeakerList, fileLabel = myAnnotation.uri, maximumSecondDifference = maximumSecondDifference, minimumSecondDuration = minimumSecondDuration)
return newList, newAnnotation
def loadAudioRTTM(sampleRTTM):
# Read in prediction data
# Data in list form, for convenient plotting
speakerList = []
# Data in Annotation form, for convenient error rate calculation
prediction = Annotation(uri=sampleRTTM)
with open(sampleRTTM, "r") as rttm:
for line in rttm:
speakerResult = line.split(' ')
index = int(speakerResult[7][-2:])
start = float(speakerResult[3])
end = start + float(speakerResult[4])
while len(speakerList) < index + 1:
speakerList.append([])
speakerList[index].append((float(speakerResult[3]),float(speakerResult[4])))
prediction[Segment(start,end)] = speakerResult[7]
return speakerList, prediction
def loadAudioTXT(sampleTXT):
# Read in prediction data
# Data in list form, for convenient plotting
speakerList = []
# Data in Annotation form, for convenient error rate calculation
prediction = Annotation(uri=sampleTXT)
with open(sampleTXT, "r") as txt:
for line in txt:
speakerResult = line.split('\t')
print(speakerResult)
if len(speakerResult) < 3:
continue
index = -1
start = float(speakerResult[0])
end = float(speakerResult[1])
duration = end - start
prediction[Segment(start,end)] = speakerResult[2]
return [], prediction
def loadAudioCSV(sampleCSV):
# Read in prediction data
df = pd.read_csv(sampleCSV)
df = df.reset_index() # make sure indexes pair with number of rows
# Data in Annotation form, for convenient error rate calculation
prediction = Annotation(uri=sampleCSV)
for i, row in df.iterrows():
index = row['Resource']
start = row['Start']
end = row['Finish']
prediction[Segment(start,end)] = index
return [], prediction
def splitIntoTimeSegments(testFile,maxDurationInSeconds=60):
waveform, sample_rate = torchaudio.load(testFile)
audioSegments = []
outOfBoundsIndex = waveform.shape[-1]
currentStart = 0
currentEnd = min(maxDurationInSeconds * sample_rate,outOfBoundsIndex)
done = False
while(not done):
waveformSegment = waveform[:,currentStart:currentEnd]
audioSegments.append(waveformSegment)
if currentEnd >= outOfBoundsIndex:
done = True
break
else:
currentStart = currentEnd
currentEnd = min(currentStart + maxDurationInSeconds * sample_rate,outOfBoundsIndex)
return audioSegments, sample_rate
def audioNormalize(waveform,sampleRate,stepSizeInSeconds = 2,dbThreshold = -50,dbTarget = -5):
print("In audioNormalize")
copyWaveform = waveform.clone().detach()
print("Waveform copy made")
transform = torchaudio.transforms.AmplitudeToDB(stype="amplitude", top_db=80)
currStart = 0
currEnd = int(min(currStart + stepSizeInSeconds * sampleRate, len(copyWaveform[0])-1))
done = False
while(not done):
copyWaveform_db = waveform[:,currStart:currEnd].clone().detach()
copyWaveform_db = transform(copyWaveform_db)
if currStart == 0:
print("First DB level calculated")
if torch.max(copyWaveform_db[0]).item() > dbThreshold:
gain = torch.min(dbTarget - copyWaveform_db[0])
adjustGain = torchaudio.transforms.Vol(gain,'db')
copyWaveform[0][currStart:currEnd] = adjustGain(copyWaveform[0][currStart:currEnd])
if len(copyWaveform_db) > 1:
if torch.max(copyWaveform_db[1]).item() > dbThreshold:
gain = torch.min(dbTarget - copyWaveform_db[1])
adjustGain = torchaudio.transforms.Vol(gain,'db')
copyWaveform[1][currStart:currEnd] = adjustGain(copyWaveform[1][currStart:currEnd])
currStart += int(stepSizeInSeconds * sampleRate)
if currStart > currEnd:
done = True
else:
currEnd = int(min(currStart + stepSizeInSeconds * sampleRate, len(copyWaveform[0])-1))
print("Waveform enhanced")
return copyWaveform
class equalizeVolume(torch.nn.Module):
def forward(self, waveform,sampleRate,stepSizeInSeconds,dbThreshold,dbTarget):
print("In equalizeVolume")
waveformDifference = audioNormalize(waveform,sampleRate,stepSizeInSeconds,dbThreshold,dbTarget)
return waveformDifference
def combineWaveforms(waveformList):
return torch.cat(waveformList,1)
def annotationToSpeakerList(myAnnotation):
tempSpeakerList = []
tempSpeakerNames = []
for speakerName in myAnnotation.labels():
speakerIndex = None
if speakerName not in tempSpeakerNames:
speakerIndex = len(tempSpeakerNames)
tempSpeakerNames.append(speakerName)
tempSpeakerList.append([])
else:
speakerIndex = tempSpeakerNames.index(speakerName)
for segmentItem in myAnnotation.label_support(speakerName):
tempSpeakerList[speakerIndex].append((segmentItem.start,segmentItem.duration))
return tempSpeakerList
def speakerListToDataFrame(speakerList):
dataList = []
for j, row in enumerate(speakerList):
for k, speakingPoint in enumerate(row):
h0 = int(speakingPoint[0]//3600)
m0 = int(speakingPoint[0]%3600//60)
s0 = int(speakingPoint[0]%60)
ms0 = int(speakingPoint[0]*1000000%1000000)
time0 = dt.time(h0,m0,s0,ms0)
dtStart = dt.datetime.combine(dt.date.today(), time0)
endPoint = speakingPoint[0] + speakingPoint[1]
h1 = int(endPoint//3600)
m1 = int(endPoint%3600//60)
s1 = int(endPoint%60)
ms1 = int(endPoint*1000000%1000000)
time1 = dt.time(h1,m1,s1,ms1)
dtEnd = dt.datetime.combine(dt.date.today(), time1)
dataList.append(dict(Task=f"Speaker {j}.{k}", Start=dtStart, Finish=dtEnd, Resource=f"Speaker {j+1}"))
df = pd.DataFrame(dataList)
return df
def removeOverlap(timeSegment,overlap):
times = []
if timeSegment.start < overlap.start:
times.append(Segment(timeSegment.start,min(overlap.start,timeSegment.end)))
if timeSegment.end > overlap.end:
times.append(Segment(max(timeSegment.start,overlap.end),timeSegment.end))
return times
def checkForOverlap(time1, time2):
overlap = time1 & time2
if overlap:
return overlap
else:
return None
def sumSegments(segmentList):
total = 0
for s in segmentList:
total += s.duration
return total
def sumTimes(myAnnotation):
return myAnnotation.get_timeline(False).duration()
def sumTimesPerSpeaker(myAnnotation):
speakerList = []
timeList = []
for speaker in myAnnotation.labels():
if speaker not in speakerList:
speakerList.append(speaker)
timeList.append(0)
timeList[speakerList.index(speaker)] += sumTimes(myAnnotation.subset([speaker]))
return speakerList, timeList
def sumMultiTimesPerSpeaker(myAnnotation):
speakerList = []
timeList = []
sList,tList = sumTimesPerSpeaker(myAnnotation)
for i,speakerGroup in enumerate(sList):
speakerSplit = speakerGroup.split('+')
for speaker in speakerSplit:
if speaker not in speakerList:
speakerList.append(speaker)
timeList.append(0)
timeList[speakerList.index(speaker)] += tList[i]
return speakerList, timeList
def annotationToDataFrame(myAnnotation):
dataList = []
speakerDict = {}
for currSpeaker in myAnnotation.labels():
if currSpeaker not in speakerDict.keys():
speakerDict[currSpeaker] = []
for currSegment in myAnnotation.subset([currSpeaker]).itersegments():
speakerDict[currSpeaker].append(currSegment)
timeSummary = {}
for key in speakerDict.keys():
if key not in timeSummary.keys():
timeSummary[key] = 0
for speakingSegment in speakerDict[key]:
timeSummary[key] += speakingSegment.duration
for key in speakerDict.keys():
for k, speakingSegment in enumerate(speakerDict[key]):
speakerName = key
startPoint = speakingSegment.start
endPoint = speakingSegment.end
h0 = int(startPoint//3600)
m0 = int(startPoint%3600//60)
s0 = int(startPoint%60)
ms0 = int(startPoint*1000000%1000000)
time0 = dt.time(h0,m0,s0,ms0)
dtStart = dt.datetime.combine(dt.date.today(), time0)
h1 = int(endPoint//3600)
m1 = int(endPoint%3600//60)
s1 = int(endPoint%60)
ms1 = int(endPoint*1000000%1000000)
time1 = dt.time(h1,m1,s1,ms1)
dtEnd = dt.datetime.combine(dt.date.today(), time1)
dataList.append(dict(Task=speakerName + f".{k}", Start=dtStart, Finish=dtEnd, Resource=speakerName))
df = pd.DataFrame(dataList)
return df, timeSummary
def annotationToSimpleDataFrame(myAnnotation):
dataList = []
speakerDict = {}
for currSpeaker in myAnnotation.labels():
if currSpeaker not in speakerDict.keys():
speakerDict[currSpeaker] = []
for currSegment in myAnnotation.subset([currSpeaker]).itersegments():
speakerDict[currSpeaker].append(currSegment)
timeSummary = {}
for key in speakerDict.keys():
if key not in timeSummary.keys():
timeSummary[key] = 0
for speakingSegment in speakerDict[key]:
timeSummary[key] += speakingSegment.duration
for key in speakerDict.keys():
for k, speakingSegment in enumerate(speakerDict[key]):
speakerName = key
startPoint = speakingSegment.start
endPoint = speakingSegment.end
dataList.append(dict(Task=speakerName + f".{k}", Start=startPoint, Finish=endPoint, Resource=speakerName))
df = pd.DataFrame(dataList)
return df, timeSummary
def calcCategories(myAnnotation,categories):
categorySlots = []
extraCategories = []
for category in categories:
categorySlots.append([])
for speaker in myAnnotation.labels():
targetCategory = None
for i, category in enumerate(categories):
if speaker in category:
targetCategory = i
if targetCategory is None:
targetCategory = len(categorySlots)
categorySlots.append([])
extraCategories.append(speaker)
for timeSegment in myAnnotation.subset([speaker]).itersegments():
categorySlots[targetCategory].append((speaker,timeSegment))
# Clean up categories
cleanCategories = []
for category in categorySlots:
newCategory = []
catSorted = copy.deepcopy(sorted(category,key=lambda cSegment: cSegment[1].start))
currID, currSegment = None, None
if len(catSorted) > 0:
currID, currSegment = catSorted[0]
for sp, segmentSlot in catSorted[1:]:
overlapTime = checkForOverlap(currSegment,segmentSlot)
if overlapTime is None:
newCategory.append((currID,currSegment))
currID = sp
currTime = segmentSlot
else:
currID = currID + "+" + sp
# Union of segments
currTime[1] = currSegment | segmentSlot
if currSegment is not None:
newCategory.append((currID,currSegment))
cleanCategories.append(newCategory)
return cleanCategories,extraCategories
def calcSpeakingTypes(myAnnotation,maxTime):
noVoice = [Segment(0,maxTime)]
oneVoice = []
multiVoice = []
for speaker in myAnnotation.labels():
timesToProcess = []
for timeSegment in myAnnotation.subset([speaker]).itersegments():
timesToProcess.append((speaker,timeSegment))
while len(timesToProcess) > 0:
currID, currSegment = timesToProcess[0]
timesToProcess.remove(timesToProcess[0])
resetCheck = False
# Check in multi
for compareID,timeSegment in multiVoice:
overlapTime = checkForOverlap(currSegment,timeSegment)
if overlapTime is None:
continue
else:
compareID.append(currID)
newTimes = removeOverlap(currSegment,timeSegment)
for i in range(len(newTimes)):
newTimes[i] = (currID,newTimes[i])
timesToProcess += newTimes
resetCheck = True
break
if resetCheck:
continue
# Check in one voice
for timeSlot in oneVoice:
tID = timeSlot[0]
tSegment = timeSlot[1]
overlapTime = checkForOverlap(currSegment,tSegment)
if overlapTime is None:
continue
else:
oneVoice.remove(timeSlot)
# Add back non overlap
newTimes = removeOverlap(tSegment,currSegment)
for i in range(len(newTimes)):
newTimes[i] = (tID,newTimes[i])
oneVoice += newTimes
# Add overlap time to multivoice
multiVoice.append(([tID,currID],overlapTime))
# Add new times back to process
newTimes = removeOverlap(currSegment,tSegment)
for i in range(len(newTimes)):
newTimes[i] = (currID,newTimes[i])
timesToProcess += newTimes
resetCheck = True
break
if resetCheck:
continue
# Add to one voice
oneVoice.append((currID,currSegment))
ovAnnotation = Annotation()
mvAnnotation = Annotation()
for currID,timeSlot in multiVoice:
currIDString = '+'.join(currID)
mvAnnotation[timeSlot] = currIDString
copyOfNo = copy.deepcopy(noVoice)
for emptySlot in noVoice:
if checkForOverlap(timeSlot,emptySlot) is None:
continue
else:
copyOfNo.remove(emptySlot)
copyOfNo += removeOverlap(emptySlot,timeSlot)
noVoice = copyOfNo
for currID,timeSlot in oneVoice:
ovAnnotation[timeSlot] = currID
copyOfNo = copy.deepcopy(noVoice)
for emptySlot in noVoice:
if checkForOverlap(timeSlot,emptySlot) is None:
continue
else:
copyOfNo.remove(emptySlot)
copyOfNo += removeOverlap(emptySlot,timeSlot)
noVoice = copyOfNo
nvAnnotation = Annotation()
for emptySlot in noVoice:
nvAnnotation[emptySlot] = "None"
return nvAnnotation, ovAnnotation, mvAnnotation
def timeToString(timeInSeconds):
if isinstance(timeInSeconds,list):
return [timeToString(t) for t in timeInSeconds]
else:
h = int(timeInSeconds//3600)
m = int(timeInSeconds%3600//60)
s = timeInSeconds%60
return f'{h:02d}::{m:02d}::{s:02.2f}' |