Add ASR CTC inference tutorial #2106
Conversation
| # ~~~~~~~~~~~~~~~~ | ||
| # | ||
|
|
||
| def download_file(url, file): |
There was a problem hiding this comment.
This might be coming from the other tutorials but I think we can simply use torch.hub.download_url_to_file.
| # | ||
| # The tokens are the possible symbols that the acoustic model can predict, | ||
| # including the blank and silent symbols. | ||
| # |
There was a problem hiding this comment.
Similar to lexicon bellow, can you show a part of the contents of token file?
| speech_url = "https://pytorch.s3.amazonaws.com/torchaudio/tutorial-assets/ctc-decoding/8461-258277-0000.flac" | ||
| speech_file = "_assets/speech.flac" | ||
| speech_url = "https://pytorch.s3.amazonaws.com/torchaudio/tutorial-assets/ctc-decoding/8461-258277-0000.wav" | ||
| speech_file = "/tmp/speech.wav" |
There was a problem hiding this comment.
Can you use torch.hub.get_dir? So that configurable default temporary directory will be selected.
Hardcoding /tmp would leave artifacts when running locally.
carolineechen
force-pushed
the
ctc-inference-tutorial
branch
from
256f9e7 to
ee9c8c9
Compare
| # We see that the transcript with the lexicon-constrained beam search | ||
| # decoder consists of real words, while the greedy decoder can predict | ||
| # incorrectly spelled words like “hundrad”. | ||
| # |
There was a problem hiding this comment.
Please run pre-commit run once again.
|
@carolineechen has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
|
@carolineechen has imported this pull request. If you are a Meta employee, you can view this diff on Phabricator. |
Summary: demonstrate usage of the CTC beam search decoder w/ lexicon constraint and KenLM support, on a LibriSpeech sample and using a pretrained wav2vec2 model rendered: https://485200-90321822-gh.circle-artifacts.com/0/docs/tutorials/asr_inference_with_ctc_decoder_tutorial.html follow-ups: - incorporate `nbest` - demonstrate customizability of different beam search parameters Pull Request resolved: pytorch#2106 Reviewed By: mthrok Differential Revision: D33340946 Pulled By: carolineechen fbshipit-source-id: 0ab838375d96a035d54ed5b5bd9ab4dc8d19adb7
Add tags to the maskedtensor tutorials so they appear when filtering by tag
demonstrate usage of the CTC beam search decoder w/ lexicon constraint and KenLM support, on a LibriSpeech sample and using a pretrained wav2vec2 model
rendered: https://485200-90321822-gh.circle-artifacts.com/0/docs/tutorials/asr_inference_with_ctc_decoder_tutorial.html
follow-ups:
nbest