sercanto/podkastomat
1
0
Fork 0
Code Issues Pull requests Actions Packages Projects Releases Wiki Activity

Support using Vosk instead of Whisper for transcripts

Browse source 
For Esperanto, Vosk is used automatically, since Whisper doesn't
support Esperanto
This commit is contained in:
Benno Lang 2025-01-23 15:50:05 +10:30
parent 62f4b7bedd
commit ff23de60bb
3 changed files with 21 additions and 10 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

5
lib/args.py
View file

@ -11,6 +11,7 @@ class Options:
filter = None filter = None
episodes = 'new' episodes = 'new'
numEpisodes = 1 numEpisodes = 1
tts = 'whisper'
def outputHelp(): def outputHelp():
print("Usage: ./process [args] [filter [age] [episodes]]") print("Usage: ./process [args] [filter [age] [episodes]]")
@ -26,6 +27,8 @@ def outputHelp():
print(" Skip generating transcripts and translations of podcast episodes") print(" Skip generating transcripts and translations of podcast episodes")
print("--verbose") print("--verbose")
print(" Output a detailed report of actions as they are being taken") print(" Output a detailed report of actions as they are being taken")
print("--vosk")
print(" Use Vosk instead of Whisper to generate transcripts")
print("") print("")
print("filter:") print("filter:")
print(" If specified, only configured podcasts which match the filter will be processed") print(" If specified, only configured podcasts which match the filter will be processed")
@ -61,6 +64,8 @@ def read():
opts.generate = False opts.generate = False
elif arg == "--verbose": elif arg == "--verbose":
opts.verbose = True opts.verbose = True
elif arg == '--vosk':
opts.tts = 'vosk'
elif arg == 'new': elif arg == 'new':
opts.episodes = 'new' opts.episodes = 'new'
elif arg == 'old': elif arg == 'old':

18
lib/files.py
View file

@ -109,7 +109,7 @@ def hasTranslation(audioFile):
translationFile = getTranslationFilename(audioFile) translationFile = getTranslationFilename(audioFile)
return os.path.isfile(translationFile) return os.path.isfile(translationFile)
def generateFromAudio(audioFile, task): def generateFromAudio(audioFile, task, options):
if task == 'transcribe': if task == 'transcribe':
newExt = 'transcript.txt' newExt = 'transcript.txt'
else: else:
@ -119,14 +119,18 @@ def generateFromAudio(audioFile, task):
langCode = getLangCode(audioFile) langCode = getLangCode(audioFile)
fileParts = audioFile.split('/') fileParts = audioFile.split('/')
fileName = fileParts.pop() fileName = fileParts.pop()
newFileName = replaceExtension(fileName, newExt)
dir = '/'.join(fileParts) dir = '/'.join(fileParts)
os.chdir(dir) os.chdir(dir)
cmd = f"whisper {fileName} --model medium --language {langCode} --task {task} --output_format vtt --fp16 False" if options.tts == 'vosk' or langCode == 'eo':
os.system(cmd) cmd = f"vosk-transcriber -i {fileName} -l {langCode} -t srt -o {newFileName}"
os.system(cmd)
else:
cmd = f"whisper {fileName} --model medium --language {langCode} --task {task} --output_format srt --fp16 False"
os.system(cmd)
# rename transcript/translation file generated # rename transcript/translation file generated
generatedFile = replaceExtension(fileName, 'vtt') generatedFile = replaceExtension(fileName, 'vtt')
newFileName = replaceExtension(fileName, newExt) os.rename(generatedFile, newFileName)
os.rename(generatedFile, newFileName)
os.chdir('../../..') os.chdir('../../..')

8
process
View file

@ -30,11 +30,13 @@ for audioFile in audioFiles:
if not files.hasTranscript(audioFile): if not files.hasTranscript(audioFile):
skipped = False skipped = False
print(f"Transcribing audio: {audioFile} ({audioLength})", flush=True) print(f"Transcribing audio: {audioFile} ({audioLength})", flush=True)
files.generateFromAudio(audioFile, 'transcribe') files.generateFromAudio(audioFile, 'transcribe', options)
if language in config['translate'] and not files.hasTranslation(audioFile): if (language in config['translate'] and not files.hasTranslation(audioFile)
and language != 'eo' and options.tts != 'vosk'
):
skipped = False skipped = False
print(f"Translating audio: {audioFile} ({audioLength})", flush=True) print(f"Translating audio: {audioFile} ({audioLength})", flush=True)
files.generateFromAudio(audioFile, 'translate') files.generateFromAudio(audioFile, 'translate', options)
if skipped and options.verbose: if skipped and options.verbose:
print(f"Skipped audio: {audioFile} ({audioLength})") print(f"Skipped audio: {audioFile} ({audioLength})")