VisualEars FastConformer FA 69M ONNX FP16

Browser-ready ONNX export for the VisualEars 69M Persian ASR model.

  • Weights: FP16
  • Public IO: float32 processed_signal, int64 processed_signal_length
  • Outputs: float32 logits, int64 encoded_lengths
  • Fixed feature frames: 2005
  • Vocab: BPE 1024 + CTC blank

Validation on the VisualEars worst-269 parity set:

  • Exact transcript parity: 269/269 = 1.0
  • Mean char similarity: 1.0
  • Argmax step agreement: 0.9997360781208762
  • ONNX nonempty rate: 267/269 = 0.9925650557620818

See export_manifest_fp16_iofp32.json and parity_full269_len_fp16_iofp32.json.

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