Reza2kn/visualears-persian-asr-16k
Viewer • Updated • 3.93M • 542 • 1
Koochik (کوچیک, "small") is the 114M teacher / flagship of the Shenava‑1 family — a FastConformer Hybrid RNNT/CTC model fine‑tuned on clean Persian data. This repository contains the CTC non-streaming INT8 quantized ONNX export for use with sherpa-onnx.
| Property | Value |
|---|---|
| Parent Model | Reza2kn/Shenava-Koochik-v1.0 |
| Architecture | FastConformer Hybrid RNNT/CTC → CTC decoder export |
| Language | Persian (Farsi) |
| Parameters | 114M |
| Vocabulary Size | 1025 tokens (BPE + blank) |
| Tokenizer | ve_tok_v4 — SentencePiece BPE‑1024 |
| Sample Rate | 16 kHz |
| Subsampling | 8× |
| Streaming | ❌ No (full-context) |
| Weight Type | INT8 (dynamic quantization) |
| Author | Reza2kn |
| Version | 1.0 |
.
├── README.md
├── model.int8.onnx # ONNX model (INT8 quantized)
└── 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