---
language:
- az
license: cc-by-nc-4.0
library_name: nemo
pipeline_tag: automatic-speech-recognition
tags:
- asr
- ctc
- fastconformer
- kenlm
- azerbaijani
---

# speakwrite-azerbaijani


Azerbaijani speech recognition for CPU: FastConformer-CTC (18 layers, d512) with a 3-pass loop refiner, product-key acoustic memory
(700 x 700 sub-keys) and a 6-gram token KenLM. Input is 16 kHz mono; output has case and `. , ? !`, with numbers written as words.

## Results

[LocalDoc/fleurs-azerbaijani-asr](https://huggingface.co/datasets/LocalDoc/fleurs-azerbaijani-asr), not used for training. Reference: `text_norm`
(numbers as words). WER and CER ignore case and punctuation. Clips with `norm_status = failed` are excluded (11 of 3977).

| split | clips | hours | no LM: WER / CER | KenLM: WER / CER |
|---|---|---|---|---|
| test | 914 | 3.20 | 16.34 / 4.75 | **10.98 / 3.57** |
| dev | 397 | 1.34 | 16.52 / 4.75 | 11.05 / 3.50 |
| train | 2655 | 9.27 | 15.82 / 4.80 | 11.30 / 3.78 |
| all | 3966 | 13.81 | 16.01 / 4.79 | 11.20 / 3.70 |

- Decoder weights (`alpha` 1.0, `beta` 2.0) were chosen on the FLEURS dev split from 8 pairs, so the test row is the one to quote.
- Sentences of the FLEURS test split were excluded from the KenLM text; dev and train sentences were not.
- Internal validation, no LM, on our own data mix: WER 13.48 / CER 6.01.

Speed, PyTorch float32 on one i7-12700K, in Docker, times faster than real time, for the acoustic model plus KenLM:

| threads | speed | measured on |
|---|---|---|
| 1 | 12.7x | an earlier checkpoint of this model, 100 clips |
| 8 | ~45x | this checkpoint, all FLEURS |

The machine had other load during the runs, so these are rough. KenLM takes about 1.5% of the time. ONNX runtimes were not measured.

## Use

Needs [balbal](https://github.com/erayyap/balbal) at commit `1be9f88b77c63fbafa7713c6e46e79847f6c390e` (Python 3.11-3.12, Linux; install
`requirements/full.txt`, then run `inference/build_native.py`). Copy this repository's `refiner.json` over balbal's
`configs/balbal-v2-manas/refiner.json`: this model's memory size differs from balbal's default, and the weights will not load otherwise.

```python
import json, sys
sys.path[:0] = ["balbal/inference", "balbal"]
from balbal.audio import read_audio
from runtime import Engine

engine = Engine("speakwrite-azerbaijani", runtime="torch", settings=json.load(open("speakwrite-azerbaijani/decoder.json")))
print(engine.transcribe(read_audio("speech.mp3"), itn=False)["text"])
```

Keep `itn=False`: balbal's number conversion is Turkish. Long audio is cut into fixed 30 s chunks.

To use the model without the KenLM, take the argmax of the CTC output instead of `Engine`: without an LM, WER on FLEURS test is 16.34. The `lm/` folder is then not needed.

## Training

About 2800 hours of Azerbaijani speech. Encoder initialized from
[nvidia/stt_en_fastconformer_ctc_large](https://huggingface.co/nvidia/stt_en_fastconformer_ctc_large), then modified and trained for 4.5 epochs
(127552 updates) with the balbal-v2-manas recipe. balbal's own weights were not used.

## License

[CC BY-NC 4.0](https://creativecommons.org/licenses/by-nc/4.0/). Non-commercial use only; credit is required in any project using this model:
"speakwrite-azerbaijani by LocalDoc" with a link to this page. For commercial use, contact [LocalDoc](https://huggingface.co/LocalDoc).

Based on `nvidia/stt_en_fastconformer_ctc_large` (CC BY 4.0, NVIDIA) and on code from balbal (Apache 2.0).