latest faster-whisper support added

[Hasan-Naseer]

• requirements.txt updated
• latest distil models are now available
• asr.py updated with new parameters for faster-whisper 1.0.3
• support for hotwords added

[RichardQin1]

hellow ，Now I want to use the hotwords feature in the latest version of FastWhisper. Which branch do you recommend using. Additionally, I attempted to use the fastwhisper model separately and employed Whisperx Dialization Pipeline may encounter errors due to model conflicts.

def load_asr_model():
    model = faster_whisper.WhisperModel("large-v3", device=global_var.MODEL_DEVICE, compute_type=global_var.MODEL_COMPUTE_TYPE)
    return model

def load_speaker_model():
    diarize_model = whisperx.DiarizationPipeline(use_auth_token='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',
                                                 device=global_var.MODEL_DEVICE)
    return diarize_model

simultaneous loading print error:

Could not load library libcudnn_cnn_infer.so.8. Error: /home/phoenix/anaconda3/envs/dataProductionTasks/bin/../lib/libcudnn_cnn_infer.so.8: undefined symbol: _ZN11nvrtcHelper4loadEb, version libcudnn_ops_infer.so.8

···
conda install pytorch==2.0.0 torchaudio==2.0.0 pytorch-cuda=11.8 -c pytorch -c nvidia

···
Can you please teach me
thank！！！

[Hasan-Naseer]

hellow ，Now I want to use the hotwords feature in the latest version of FastWhisper. Which branch do you recommend using. Additionally, I attempted to use the fastwhisper model separately and employed Whisperx Dialization Pipeline may encounter errors due to model conflicts.

def load_asr_model():
    model = faster_whisper.WhisperModel("large-v3", device=global_var.MODEL_DEVICE, compute_type=global_var.MODEL_COMPUTE_TYPE)
    return model

def load_speaker_model():
    diarize_model = whisperx.DiarizationPipeline(use_auth_token='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',
                                                 device=global_var.MODEL_DEVICE)
    return diarize_model

simultaneous loading print error:

Could not load library libcudnn_cnn_infer.so.8. Error: /home/phoenix/anaconda3/envs/dataProductionTasks/bin/../lib/libcudnn_cnn_infer.so.8: undefined symbol: _ZN11nvrtcHelper4loadEb, version libcudnn_ops_infer.so.8

··· conda install pytorch==2.0.0 torchaudio==2.0.0 pytorch-cuda=11.8 -c pytorch -c nvidia

··· Can you please teach me thank！！！

Hey. It depends on whether you're cloning the repo and building it to run or just installing using pip.

For cloning the repo:
open the working directory of your repo
run

git pull origin release/latest-faster-whisper-version

This should give you the required branch. Now when you install the dependencies from requirements.txt you should have everything up to date.

Alternatively, and this is the method I recommend, use it through pip.

Install the fork https://github.com/Hasan-Naseer/whisperX.git. The main branch has the commits for latest faster-whisper (hotwords) support and has prnt_segments and prnt_duration parameters for better verbose and debugging.

Make sure you're in the directory environment with whisperX installed
run

pip uninstall faster-whisper whisperx

pip install git+https://github.com/Hasan-Naseer/whisperX.git

This should give you the latest commit with all the features.

PS. Your error seems to be related to a dependency problem in the "libcudnn" library. It is seemingly unrelated to faster-whisper or whisperx. Please look into a solution for that.

[zckrs]

@m-bain any news about merge this ?