Cutlass — English word-timing model (ONNX)

ONNX export of torchaudio's WAV2VEC2_ASR_BASE_960H — wav2vec 2.0 base fine-tuned on LibriSpeech 960 h, published by the wav2vec 2.0 authors (fairseq) under the MIT License and redistributed under the same license. The Cutlass video editor downloads it on first use to place word times in a transcript (forced alignment).

Changes from the original

Exported to ONNX (opset 17). Input audio (1 × samples, raw 16 kHz mono), output log_probs (log-softmax).
wav2vec2_en.labels.json: one character per output column (- = CTC blank, | = word boundary).

File	sha256
`wav2vec2_en.onnx`	`f54db4b70e3a13810d51c17d28a0e97dfb7dafd93ebe6d9bd812b988b804a046`
`wav2vec2_en.labels.json`	`22ccf434ba3f785b42bcfee075ed5ad81b554213ecabf0bfb3a46d1d431f1b25`

License: MIT, see LICENSE (Copyright (c) Facebook, Inc. and its affiliates).

Downloads last month: -; Downloads are not tracked for this model. How to track

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support