| --- |
| language: |
| - fa |
| license: cc-by-nc-4.0 |
| library_name: nemo |
| pipeline_tag: automatic-speech-recognition |
| base_model: |
| - Reza2kn/Shenava-Koochik-v1.0 |
| base_model_relation: finetune |
| tags: |
| - automatic-speech-recognition |
| - speech |
| - persian |
| - farsi |
| - fastconformer |
| - ctc |
| - streaming |
| - on-device |
| - shenava |
| - shenava-1 |
| - visualears |
| - rnnt |
| - nemo |
| - distillation |
| metrics: |
| - wer |
| - cer |
| datasets: |
| - Reza2kn/visualears-persian-asr-16k |
| - Reza2kn/visualears-golden-6669 |
| - Reza2kn/fleurs-fa-benchmark |
| --- |
| |
| # Shenava — Rizeh v1.0 (32M) · Persian streaming ASR |
|
|
| **Rizeh** (ریزه, “tiny”) is the 32M mid‑tier of the **Shenava‑1** family — a FastConformer Hybrid RNNT/CTC model distilled (logit + feature KD) from the [Koochik v1.0 114M teacher](https://huggingface.co/Reza2kn/Shenava-Koochik-v1.0). CTC head deployed. fp32 NeMo source; quants in own repos below. |
|
|
| ## The Shenava‑1 family |
|
|
| A knowledge‑distillation cascade of on‑device Persian ASR models — one teacher distilled down to a 6.9M student. This model is one member; its siblings: |
|
|
| - [`Reza2kn/Shenava-Koochik-v1.0`](https://huggingface.co/Reza2kn/Shenava-Koochik-v1.0) — **Koochik v1.0** (114M) · teacher / flagship — on-device WER record |
| - [`Reza2kn/Shenava-Rizeh-v1.0`](https://huggingface.co/Reza2kn/Shenava-Rizeh-v1.0) — **Rizeh v1.0** (32M) · mid-tier student ◀ **this model (or its parent)** |
| - [`Reza2kn/Shenava-Rizeh-Pizeh-v1.0`](https://huggingface.co/Reza2kn/Shenava-Rizeh-Pizeh-v1.0) — **Rizeh Pizeh v1.0** (6.9M) · tiniest — real-time on a 2015 Cortex-A7 |
|
|
| ## Benchmark — fair WER/CER |
|
|
| ITN + Persian‑digit normalizer (the [double‑benchmark](https://huggingface.co/spaces/Reza2kn/persian-asr-double-benchmark) convention), decoded @ `att_context_size=[70,13]`. |
|
|
| | 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% | |
|
|
| Koochik v1.0 is **#2 on the public double‑benchmark, behind only cloud Gemini** — the best on‑device Persian ASR, beating a 1.5B Whisper‑Persian by >2× WER at 1/13 the size. |
|
|
| ## Quantized formats (own repos) |
|
|
| - [`Shenava-Rizeh-v1.0-ONNX-fp16`](https://huggingface.co/Reza2kn/Shenava-Rizeh-v1.0-ONNX-fp16) |
| - [`Shenava-Rizeh-v1.0-CoreML-fp16`](https://huggingface.co/Reza2kn/Shenava-Rizeh-v1.0-CoreML-fp16) |
|
|
| 32M, d_model 256 / 16 layers, ×8 subsampling, multi‑latency `[[70,13],[70,6],[70,1],[70,0]]`. |
| |
| Tokenizer: ve_tok_v4 (SentencePiece BPE‑1024 +blank, digit/punct/«»‑aware). Numbers are **spoken‑form**; apply ITN at display for digits. Part of [VisualEars / Shenava](https://shenava.app). |
| |
