You can deploy this model and start earning money today!

mah92/
sherpa-onnx-nemo-ctc-fa-shenava-koochik-v1.0-non-streaming-2026-06-26

Name: sherpa-onnx-nemo-ctc-fa-shenava-koochik-v1.0-non-streaming-2026-06-26
Availability: InStock

Views

No views yet

Shenava — Koochik v1.0 (114M) · CTC Non-Streaming · ONNX (FP32)

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 with ve_tok_v4. This repository contains the CTC non-streaming FP32 ONNX export for use with sherpa-onnx.

The original NeMo checkpoint is at Reza2kn/Shenava-Koochik-v1.0. This export uses the CTC decoder head only, without streaming cache support — processes full audio at once for maximum accuracy.

The Shenava‑1 family

Member	Size	HF Repo
Koochik v1.0 (teacher)	114M	`Reza2kn/Shenava-Koochik-v1.0`
Rizeh v1.0 (mid student)	32M	`Reza2kn/Shenava-Rizeh-v1.0`
Rizeh Pizeh v1.0 (tiny student)	6.9M	`Reza2kn/Shenava-Rizeh-Pizeh-v1.0`

Benchmark — fair WER/CER

ITN + Persian‑digit normalizer (the double‑benchmark convention).

Member	golden‑6669 WER	CER	FLEURS‑fa WER	CER
Koochik v1.0 (114M)	7.49%	2.30%	10.64%	3.79%
Rizeh v1.0 (32M)	12.11%	3.94%	14.45%	5.10%
Rizeh Pizeh v1.0 (6.9M)	24.55%	8.89%	26.95%	10.22%

Note: Benchmark numbers are from the original NeMo model. The ONNX CTC export may show slight degradation.

Model Details

Property	Value
Parent Model	`Reza2kn/Shenava-Koochik-v1.0`
Model Type	EncDecHybridRNNTCTCBPEModel (CTC head only)
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	FP32 (full precision)
Author	Reza2kn
Version	1.0

Files

.
├── README.md
├── model.onnx        # ONNX model (FP32)
└── tokens.txt        # Token vocabulary (1025 tokens)

Usage

bash

sherpa-onnx-offline \
  --tokens=tokens.txt \
  --encoder=model.onnx \
  --num-threads=4 \
  /path/to/test.wav

Performance

Non-streaming variant processes full audio context for maximum accuracy. Recommended for offline/batch processing where latency is not critical.

Conversion

Exported from a NeMo checkpoint.

License

CC-BY-NC 4.0 (inherited from the parent model)

mah92/sherpa-onnx-nemo-ctc-fa-shenava-koochik-v1.0-non-streaming-2026-06-26