Nos ASR Wav2vec2 Xls R 300m Gl Onnx

--- license: apache-2.0 language: - gl tags: - onnx - onnx-asr - wav2vec2 - automatic-speech-recognition base_model: - proxectonos/Nos_ASR-wav2vec2-xls-r-300m-gl --- # Nos_ASR-wav2vec2-xls-r-300m-gl (ONNX) ONNX export of [proxectonos/Nos_ASR-wav2vec2-xls-r-300m-gl](https://huggingface.co/proxectonos/Nos_ASR-wav2vec2-xls-r-300m-gl), a Galician wav2vec2 XLS-R 300M CTC ASR model by Proxecto Nós, for use with [onnx-asr](https://github.com/istupakov/onnx-asr) (`wav2vec2-ctc` model type). Per-utterance zero-mean/unit-variance normalization (`do_normalize=True` in the original feature extractor) is baked into the ONNX graph, masked by `input_lengths` for correct behavior with padded/batched input, so the model works with onnx-asr's plain `identity` preprocessor (raw 16kHz waveform in). ## Usage ```py import onnx_asr model = onnx_asr.load_model("wav2vec2-ctc", "path/to/downloaded/model") # or, once registered upstream with a friendly name: # model = onnx_asr.load_model("OpenVoiceOS/Nos_ASR-wav2vec2-xls-r-300m-gl-onnx") print(model.recognize("test.wav")) ``` ## Files * `model.onnx` / `model.onnx.data` — fp32 ONNX graph (inputs: `input_values` (batch, samples) float32, `input_lengths` (batch,) int64; output: `logprobs` (batch, frames, vocab) float32 log-softmax). * `vocab.txt` — CTC vocabulary in onnx-asr's `token id` format (word-delimiter → `▁`, pad token → `<blk>`). * `config.json` — `{"model_type": "wav2vec2-ctc", "subsampling_factor": 320}`. No int8 quantized variant is included yet — `onnxruntime.quantization` does not currently support the `torch.onnx` dynamo-exported graph for this architecture. ## License Apache 2.0, inherited from the base model.