dekelabe's picture
Upload README.md with huggingface_hub
31785ff verified
|
Raw
History Blame Contribute Delete
569 Bytes
metadata
license: apache-2.0
language:
  - he
tags:
  - wav2vec2
  - ctc
  - onnx
  - int8

chords-align-he-int8

Int8 dynamic-quantized (MatMul-only) ONNX export of imvladikon/wav2vec2-large-xlsr-53-hebrew, used by the chords-app alignment follower: on-device forced alignment of sung Hebrew against known lyrics (position tracking, not transcription).

Input: input_values — float32 [1, samples], 16kHz mono, per-window z-normalized. Output: logits — [1, frames, vocab] at 50 fps. Vocab in vocab.json.