Audio Transcript Task not work!

[Giak1234]

Good morning, I am trying some tests with IPEAD, which works, except for transcribing audio files. Any suggestions to solve the problem @lfcnassif ? Tnk

[lfcnassif]

What IPED version are you using? Have you changed the default Vosk transcription algorithm? Could you share a processing log?Could you provide a step by step to reproduce the problem?

[Giak1234]

In order: Thanks Luis ... the version is 4.1.5 ... i did not perform any setup ... can you guide me?

IPED.log

[lfcnassif]

You simply need to set enableAudioTranscription = true into IPEDConfig.txt. The language is taken from locale into LocalConfig.txt. The default Vosk algorithm is not good, you may want to change it, see the steps on #1822.

[Giak1234]

Luis ... use a PC with proxy (error log) ... is possible bypass the connection to https://huggingface.co ?

Error.txt

[Giak1234]

I'm settings (decomment) only: implementationClass = iped.engine.task.transcript.Wav2Vec2TranscriptTask and in Local Wav2Vec2TranscriptTask options or huggingFaceModel = jonatasgrosman/wav2vec2-xls-r-1b-english or huggingFaceModel = jonatasgrosman/wav2vec2-large-xlsr-53-english

[Giak1234]

I forgot where (into AudioTranscriptConfig.txt) ... i didn't download the Vosk file (https://alphacephei.com/vosk/models) :-(

[lfcnassif]

Luis ... use a PC with proxy (error log) ... is possible bypass the connection to https://huggingface.co ?

Error.txt

If you can't disable your proxy, you should configure it in your command line before running IPED: https://stackoverflow.com/questions/11726881/how-to-set-an-http-proxy-in-python-2-7

[lfcnassif]

I forgot where (into AudioTranscriptConfig.txt) ... i didn't download the Vosk file (https://alphacephei.com/vosk/models) :-(

We already distribute Vosk models for English and Portuguese. If you want to transcribe a different language, you should download it manually from the official Vosk page and put it into the appropriate folder into iped/models folder.

[Giak1234]

Luis good morning and thank you for your time! I did more tests by adding an AD1 with audios (m4a, mp3 and oga) in it and with my current setup I got both English and Italian language transcription but only Opus files (Audio:xmpDM:audioSampleRate 16.000) while mp3 files (Audio:xmpDM:audioSampleRate 48.000) nothing! I will summarize my current file setup:

IPEDConfig.txt ---- enableAudioTranscription = true LocalConfig.txt ---- locale = en-IT

In the conf\AudioTranscriptConfig.txt file

AudioTranscriptConfig.txt

I left everything default except uncommenting the two entries huggingFaceModel = jonatasgrosman/wav2vec2-large-xlsr-53-italian and huggingFaceModel = jonatasgrosman/wav2vec2-xls-r-1b-italian ... in models\vosk I installed the file downloaded from https://alphacephei.com/vosk/models/vosk-model-it-0.22.zip and renamed it-IT. What setup should I follow to transcribe other audio file extensions? I take this opportunity to ask you about the setup to transcribe also from video files in the various formats ... thank you very much for your patience.

[lfcnassif]

What setup should I follow to transcribe other audio file extensions? I take this opportunity to ask you about the setup to transcribe also from video files in the various formats

In AudioTranscriptConfig.txt, just set mimesToProcess = audio; video so you will transcribe any audio and video formats.

LocalConfig.txt ---- locale = en-IT

This setup is wrong, use en or it-IT.

I left everything default except uncommenting the two entries huggingFaceModel = jonatasgrosman/wav2vec2-large-xlsr-53-italian and huggingFaceModel = jonatasgrosman/wav2vec2-xls-r-1b-italian ...

Just uncomment one of them, if you uncomment both, the last one overwrites the first. Anyway this configuration only works if you also uncomment: implementationClass = iped.engine.task.transcript.Wav2Vec2TranscriptTask

Otherwise the worse Vosk algorithm will be used.

in models\vosk I installed the file downloaded from https://alphacephei.com/vosk/models/vosk-model-it-0.22.zip and renamed it-IT

This is needed just if you really want to use the worse Vosk algorithm, but actually you should unzip the downloaded file and put it in a folder named models\vosk\it-IT

[Giak1234]

In AudioTranscriptConfig.txt, just set mimesToProcess = audio; video so you will transcribe any audio and video formats.

done!

This setup is wrong, use en or it-IT.

not work en-IT only en or it-IT!

Just uncomment one of them, if you uncomment both, the last one overwrites the first. Anyway this configuration only works if you also uncomment: implementationClass = iped.engine.task.transcript.Wav2Vec2TranscriptTask

if I uncomment this line, over this implementationClass = iped.engine.task.transcript.VoskTranscriptTask, i get many errors (attached file) error-en-IT.txt

This is needed just if you really want to use the worse Vosk algorithm, but actually you should unzip the downloaded file and put it in a folder named models\vosk\it-IT

already done!

[lfcnassif]

if I uncomment this line, over this implementationClass = iped.engine.task.transcript.VoskTranscriptTask, i get many errors (attached file) error-en-IT.txt

Seems a network or proxy issue to me. I'll be able to test in a couple of hours.

[Giak1234]

Thanks :-)

[lfcnassif]

Sorry for my delay, just tested implementationClass = iped.engine.task.transcript.Wav2Vec2TranscriptTask with the italian models, it worked fine. You should see something similar to this in your terminal:

Above means the model was downloaded fine. As I said, I think you are having a proxy issue.

If you can't disable your proxy, you should configure it in your command line before running IPED: https://stackoverflow.com/questions/11726881/how-to-set-an-http-proxy-in-python-2-7

As I suggested above, try to set the proxy on your command line before running IPED:

set http_proxy=http://proxy.myproxy.com
set https_proxy=https://proxy.myproxy.com