What is it • Setup • Usage • Multilingual • Contribute • More examples

Whisper-Based Automatic Speech Recognition (ASR) with improved timestamp accuracy using forced alignment.

This repository refines the timestamps of openAI's Whisper model via forced aligment with phoneme-based ASR models (e.g. wav2vec2.0), multilingual use-case.

Whisper is an ASR model developed by OpenAI, trained on a large dataset of diverse audio. Whilst it does produces highly accurate transcriptions, the corresponding timestamps are at the utterance-level, not per word, and can be inaccurate by several seconds.

Phoneme-Based ASR A suite of models finetuned to recognise the smallest unit of speech distinguishing one word from another, e.g. the element p in "tap". A popular example model is wav2vec2.0.

Forced Alignment refers to the process by which orthographic transcriptions are aligned to audio recordings to automatically generate phone level segmentation.

pip install git+https://github.com/m-bain/whisperx.git

If already installed, update package to most recent commit

pip install git+https://github.com/m-bain/whisperx.git --upgrade

If wishing to modify this package, clone and install in editable mode:

$ git clone https://github.com/m-bain/whisperX.git
$ cd whisperX
$ pip install -e .

You may also need to install ffmpeg, rust etc. Follow openAI instructions here https://github.com/openai/whisper#setup.

Run whisper on example segment (using default params)

whisperx examples/sample01.wav

For increased timestamp accuracy, at the cost of higher gpu mem, use bigger models e.g.

whisperx examples/sample01.wav --model large.en --align_model WAV2VEC2_ASR_LARGE_LV60K_960H

Result using WhisperX with forced alignment to wav2vec2.0 large:

Compare this to original whisper out the box, where many transcriptions are out of sync:

The phoneme ASR alignment model is language-specific, for tested languages these models are automatically picked from torchaudio pipelines or huggingface. Just pass in the --language code, and use the whisper --model large.

Currently default models provided for {en, fr, de, es, it, ja, zh, nl, uk, pt}. If the detected language is not in this list, you need to find a phoneme-based ASR model from huggingface model hub and test it on your data.

whisperx --model large --language de examples/sample_de_01.wav

See more examples in other languages here.

import whisperx

device = "cuda" 
audio_file = "audio.mp3"

# transcribe with original whisper
model = whisperx.load_model("large", device)
result = model.transcribe(audio_file)

# load alignment model and metadata
model_a, metadata = whisperx.load_align_model(language_code=result["language"], device=device)

# align whisper output
result_aligned = whisperx.align(result["segments"], model_a, metadata, audio_file, device)

print(result["segments"]) # before alignment

print(result_aligned["segments"]) # after alignment
print(result_aligned["word_segments"]) # after alignment

In addition to forced alignment, the following two modifications have been made to the whisper transcription method:

1. --condition_on_prev_text is set to False by default (reduces hallucination)

2. Clamping segment end_time to be at least 0.02s (one time precision) later than start_time (prevents segments with negative duration)

• Not thoroughly tested, especially for non-english, results may vary -- please post issue to let me know the results on your data
• Whisper normalises spoken numbers e.g. "fifty seven" to arabic numerals "57". Need to perform this normalization after alignment, so the phonemes can be aligned. Currently just ignores numbers.
• Assumes the initial whisper timestamps are accurate to some degree (within margin of 2 seconds, adjust if needed -- bigger margins more prone to alignment errors)
• Hacked this up quite quickly, there might be some errors, please raise an issue if you encounter any.

If you are multilingual, a major way you can contribute to this project is to find phoneme models on huggingface (or train your own) and test them on speech for the target language. If the results look good send a merge request and some examples showing its success.

The next major upgrade we are working on is whisper with speaker diarization, so if you have any experience on this please share.

[x] Multilingual init done

[x] Subtitle .ass output done

[x] Automatic align model selection based on language detection done

[x] Python usage done

[ ] Incorporating word-level speaker diarization

[ ] Inference speedup with batch processing

Contact maxbain[at]robots[dot]ox[dot]ac[dot]uk for business things.

Of course, this is mostly just a modification to openAI's whisper. As well as accreditation to this PyTorch tutorial on forced alignment

@misc{bain2022whisperx,
  author = {Bain, Max},
  title = {WhisperX},
  year = {2022},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished = {\url{https://github.com/m-bain/whisperX}},
}

as well as the whisper paper,

@article{radford2022robust,
  title={Robust speech recognition via large-scale weak supervision},
  author={Radford, Alec and Kim, Jong Wook and Xu, Tao and Brockman, Greg and McLeavey, Christine and Sutskever, Ilya},
  journal={arXiv preprint arXiv:2212.04356},
  year={2022}
}

and any alignment model used, e.g. wav2vec2.0.

@article{baevski2020wav2vec,
  title={wav2vec 2.0: A framework for self-supervised learning of speech representations},
  author={Baevski, Alexei and Zhou, Yuhao and Mohamed, Abdelrahman and Auli, Michael},
  journal={Advances in Neural Information Processing Systems},
  volume={33},
  pages={12449--12460},
  year={2020}
}