Reza2kn/visualears-persian-asr-16k
Viewer β’ Updated β’ 3.93M β’ 540 β’ 1
Rizeh Pizeh (Ψ±ΫΨ²Ω ΩΎΫΨ²Ω, "tiny grain") is the 6.9M tiniest student of the Shenavaβ1 family β capable of real-time inference on a 2015 Cortex-A7. This repository contains the CTC streaming INT8 quantized ONNX export for use with sherpa-onnx.
| Property | Value |
|---|---|
| Parent Model | Reza2kn/Shenava-Rizeh-Pizeh-v1.0 |
| Model Type | EncDecCTCModelBPE (pure CTC β no transducer head) |
| Architecture | FastConformer Pure CTC |
| Language | Persian (Farsi) |
| Parameters | 6.9M |
| Vocabulary Size | 1025 tokens (BPE + blank) |
| Tokenizer | ve_tok_v4 β SentencePiece BPEβ1024 |
| Sample Rate | 16 kHz |
| Subsampling | 8Γ |
| Streaming | β Yes (with cache support) |
| Weight Type | INT8 (dynamic quantization) |
| Author | Reza2kn |
| Version | 1.0 |
.
βββ README.md
βββ model.int8.onnx # ONNX model (INT8 quantized)
βββ tokens.txt # Token vocabulary (1025 tokens)
sherpa-onnx-offline \
--tokens=tokens.txt \
--encoder=model.int8.onnx \
--num-threads=4 \
/path/to/test.wav
Exported from a NeMo checkpoint and quantized dynamically with ONNX Runtime.
CC-BY-NC 4.0 (inherited from the parent model)
Base model
nvidia/stt_fa_fastconformer_hybrid_large