search

Marksdo / Whisper

Batch Local Transcribe Audio/Movie To Text With Whisper AI Model. Keep Privacy Safe!
https://whisper.marksdo.com
8 stars 1 forks source link

Diarization Implementation #23

Open CoDev-BitByBit-Commit opened 4 months ago

CoDev-BitByBit-Commit commented 4 months ago
  1. Diarization should be integrated into the macOS app!
  2. The provided instructions fail to consider the requirements of packages specific versions to work properly - I had to install python 3.9 in a Conda env to make it work - perhaps, as is not using Metal is going terribly slow (still running), better than manually I guess.
  3. Also, the token/API is barely described as a requirement to have it working.
CoDev-BitByBit-Commit commented 4 months ago

...lib/python3.9/site-packages/pyannote/audio/core/io.py:43: UserWarning: torchaudio._backend.set_audio_backend has been deprecated. With dispatcher enabled, this function is no-op. You can remove the function call. torchaudio.set_audio_backend("soundfile")

Marksdo commented 4 months ago

1.Diarization should be integrated into the macOS app!

Pyannote can't be embed in commercial apps

2.The provided instructions fail to consider the requirements of packages specific versions to work properly - I had to install python 3.9 in a Conda env to make it work - perhaps, as is not using Metal is going terribly slow (still running), better than manually I guess.

Yes, it's slow in mac devices as you mentioned it not use Metal or CUDA

  1. Also, the token/API is barely described as a requirement to have it working.

    Thanks for info this, I will upgrade the requirement guide in next version

Pyannote is best open source diarization project, But current has the issue as u described. Slow and can't direct embed in other app

💡 You can start Whisper Mate task then invoke the pyannote script in parallel execution.

CoDev-BitByBit-Commit commented 4 months ago

1.Diarization should be integrated into the macOS app!

Pyannote can't be embed in commercial apps

2.The provided instructions fail to consider the requirements of packages specific versions to work properly - I had to install python 3.9 in a Conda env to make it work - perhaps, as is not using Metal is going terribly slow (still running), better than manually I guess.

Yes, it's slow in mac devices as you mentioned it not use Metal or CUDA

not only slow, but also I would recommend a better documentation regarding the instructions - to set a new virtual env for this, I had to do this to get it to work on python 3.9 - this is not the latest release

  1. Also, the token/API is barely described as a requirement to have it working.

Thanks for info this, I will upgrade the requirement guide in next version

Pyannote is best open source diarization project, But current has the issue as u described. Slow and can't direct embed in other app

💡 You can start Whisper Mate task then invoke the pyannote script in parallel execution.

CoDev-BitByBit-Commit commented 4 months ago

not only slow, but also I would recommend a better documentation regarding the instructions - to set a new virtual env for this, etc... I had to do this to get it to work on python 3.9 - this is not the latest release

Marksdo commented 4 months ago

not only slow, but also I would recommend a better documentation regarding the instructions - to set a new virtual env for this, I had to do this to get it to work on python 3.9 - this is not the latest release

Get it, I will add an guide how to set it

My env is Py3.11.7 it can run pyannote.py to get result, maybe other py libs conflict