def lrs2pretrain_max_inplen_checker():
maxInpLen = 0
numWords = args["PRETRAIN_NUM_WORDS"]
for root, dirs, files in os.walk(args["DATA_DIRECTORY"] + "/pretrain"):
for file in files:
if file.endswith(".mp4"):
audioFile = os.path.join(root, file[:-4]) + ".wav"
targetFile = os.path.join(root, file[:-4]) + ".txt"
with open(targetFile, "r") as f:
lines = f.readlines()
lines = [line.strip() for line in lines]
trgt = lines[0][7:]
words = trgt.split(" ")
if len(words) <= numWords:
if len(trgt) + 1 > 256:
print("Max target length reached. Exiting")
exit()
sampFreq, audio = wavfile.read(audioFile)
inpLen = (len(audio) - 640) // 160 + 1
inpLen = int(np.ceil(inpLen / 4)) * 4
reqLen = req_input_length(trgt) + 1
if reqLen * 4 > inpLen:
inpLen = reqLen * 4
if inpLen > maxInpLen:
maxInpLen = inpLen
else:
nWords = np.array([
" ".join(words[i:i + numWords])
for i in range(len(words) - numWords + 1)
])
nWordLens = np.array([len(nWord) + 1 for nWord in nWords
]).astype(np.float)
nWordLens[nWordLens > 256] = -np.inf
if np.all(nWordLens == -np.inf):
print("Max target length reached. Exiting")
exit()
nWords = nWords[nWordLens > 0]
for ix in range(len(nWords)):
trgt = nWords[ix]
audioStartTime = float(lines[4 + ix].split(" ")[1])
audioEndTime = float(lines[4 + ix + numWords -
1].split(" ")[2])
inpLen = (
(int(sampFreq * audioEndTime) -
int(sampFreq * audioStartTime)) - 640) // 160 + 1
if inpLen < 4:
inpLen = 4
inpLen = int(np.ceil(inpLen / 4)) * 4
reqLen = req_input_length(trgt) + 1
if reqLen * 4 > inpLen:
inpLen = reqLen * 4
if inpLen > maxInpLen:
maxInpLen = inpLen
print(maxInpLen)
return
def lrs2pretrain_max_inplen_checker():
maxInpLen = 0
numWords = args["PRETRAIN_NUM_WORDS"]
for root, dirs, files in os.walk(args["DATA_DIRECTORY"] + "/pretrain"):
for file in files:
if file.endswith(".mp4"):
visualFeaturesFile = os.path.join(root, file[:-4]) + ".npy"
targetFile = os.path.join(root, file[:-4]) + ".txt"
with open(targetFile, "r") as f:
lines = f.readlines()
lines = [line.strip() for line in lines]
trgt = lines[0][7:]
words = trgt.split(" ")
if len(words) <= numWords:
if len(trgt) + 1 > 256:
print("Max target length reached. Exiting")
exit()
visualFeatures = np.load(visualFeaturesFile)
inpLen = len(visualFeatures)
reqLen = req_input_length(trgt) + 1
if reqLen > inpLen:
inpLen = reqLen
if inpLen > maxInpLen:
maxInpLen = inpLen
else:
nWords = np.array([
" ".join(words[i:i + numWords])
for i in range(len(words) - numWords + 1)
])
nWordLens = np.array([len(nWord) + 1 for nWord in nWords
]).astype(np.float)
nWordLens[nWordLens > 256] = -np.inf
if np.all(nWordLens == -np.inf):
print("Max target length reached. Exiting")
exit()
nWords = nWords[nWordLens > 0]
for ix in range(len(nWords)):
trgt = nWords[ix]
videoStartTime = float(lines[4 + ix].split(" ")[1])
videoEndTime = float(lines[4 + ix + numWords -
1].split(" ")[2])
inpLen = int(
np.ceil(args["VIDEO_FPS"] * videoEndTime) -
np.floor(args["VIDEO_FPS"] * videoStartTime))
reqLen = req_input_length(trgt) + 1
if reqLen > inpLen:
inpLen = reqLen
if inpLen > maxInpLen:
maxInpLen = inpLen
print(maxInpLen)
return
def lrs2main_max_inplen_checker():
maxInpLen = 0
for root, dirs, files in os.walk(args["DATA_DIRECTORY"] + "/main"):
for file in files:
if file.endswith(".mp4"):
audioFile = os.path.join(root, file[:-4]) + ".wav"
visualFeaturesFile = os.path.join(root, file[:-4]) + ".npy"
targetFile = os.path.join(root, file[:-4]) + ".txt"
with open(targetFile, "r") as f:
trgt = f.readline().strip()[7:]
sampFreq, audio = wavfile.read(audioFile)
audInpLen = (len(audio) - 640) // 160 + 1
visualFeatures = np.load(visualFeaturesFile)
vidInpLen = len(visualFeatures)
if vidInpLen >= audInpLen / 4:
inpLen = vidInpLen
else:
inpLen = np.ceil(audInpLen / 4)
reqLen = req_input_length(trgt) + 1
if reqLen > inpLen:
inpLen = reqLen
if inpLen > maxInpLen:
maxInpLen = inpLen
print(maxInpLen)
return
def req_input_length_checker():
strings = ["WORKS FOR DOODEE OOTY ASSAM~", " ~", "NOOOOOOOOOO~", "IT'S THAT SIMPLE~"]
for n in range(len(strings)):
trgt = list()
for i in range(len(strings[n])):
char = strings[n][i]
if char == "~":
ix = args["CHAR_TO_INDEX"]["<EOS>"]
else:
ix = args["CHAR_TO_INDEX"][char]
trgt.append(ix)
print(req_input_length(trgt))
return
def lrs2main_max_inplen_checker():
maxInpLen = 0
for root, dirs, files in os.walk(args["DATA_DIRECTORY"] + "/main"):
for file in files:
if file.endswith(".mp4"):
visualFeaturesFile = os.path.join(root, file[:-4]) + ".npy"
targetFile = os.path.join(root, file[:-4]) + ".txt"
with open(targetFile, "r") as f:
trgt = f.readline().strip()[7:]
visualFeatures = np.load(visualFeaturesFile)
inpLen = len(visualFeatures)
reqLen = req_input_length(trgt) + 1
if reqLen > inpLen:
inpLen = reqLen
if inpLen > maxInpLen:
maxInpLen = inpLen
print(maxInpLen)
return
def lrs2main_max_inplen_checker():
maxInpLen = 0
for root, dirs, files in os.walk(args["DATA_DIRECTORY"] + "/main"):
for file in files:
if file.endswith(".mp4"):
audioFile = os.path.join(root, file[:-4]) + ".wav"
targetFile = os.path.join(root, file[:-4]) + ".txt"
with open(targetFile, "r") as f:
trgt = f.readline().strip()[7:]
sampFreq, audio = wavfile.read(audioFile)
inpLen = (len(audio) - 640) // 160 + 1
inpLen = int(np.ceil(inpLen / 4)) * 4
reqLen = req_input_length(trgt) + 1
if reqLen * 4 > inpLen:
inpLen = reqLen * 4
if inpLen > maxInpLen:
maxInpLen = inpLen
print(maxInpLen)
return
