---
language:
- az
license: apache-2.0
library_name: transformers
pipeline_tag: automatic-speech-recognition
tags:
- whisper
- azerbaijani
- asr
- speech
- fine-tuned
base_model: openai/whisper-small
datasets:
- LocalDoc/azerbaijani_asr
- LocalDoc/fleurs-azerbaijani-asr
metrics:
- wer
- cer
model-index:
- name: azerbaijani-whisper-small
  results:
  - task:
      type: automatic-speech-recognition
      name: Speech Recognition
    dataset:
      type: LocalDoc/fleurs-azerbaijani-asr
      name: FLEURS Azerbaijani
      split: test
    metrics:
    - type: wer
      value: 20.54
      name: WER
    - type: cer
      value: 5.72
      name: CER
---

# Azerbaijani Whisper Small

Fine-tuned [openai/whisper-small](https://huggingface.co/openai/whisper-small) for Azerbaijani automatic speech recognition.

## Performance

| Model | Params | WER | CER |
|-------|--------|-----|-----|
| whisper-small (baseline) | 242M | 52.17% | 14.52% |
| whisper-medium (baseline) | 769M | 34.54% | 9.00% |
| whisper-large-v3 (baseline) | 1543M | 21.00% | 5.51% |
| **azerbaijani-whisper-small** | **242M** | **20.54%** | **5.72%** |

This model achieves **better quality than whisper-large-v3** while being **6x smaller**.

Evaluated on [FLEURS Azerbaijani](https://huggingface.co/datasets/LocalDoc/fleurs-azerbaijani-asr) test set.

## Usage

```bash
pip install --upgrade transformers
```

```python
import torch
import librosa
from transformers import WhisperProcessor, WhisperForConditionalGeneration
import soundfile as sf
import numpy as np

processor = WhisperProcessor.from_pretrained("LocalDoc/azerbaijani-whisper-small")
model = WhisperForConditionalGeneration.from_pretrained("LocalDoc/azerbaijani-whisper-small")

audio, sr = sf.read("audio.wav")

if len(audio.shape) > 1:
    audio = audio.mean(axis=1)

audio = librosa.resample(np.asarray(audio, dtype=np.float32), orig_sr=sr, target_sr=16000)
sr = 16000

inputs = processor(audio, sampling_rate=sr, return_tensors="pt")
forced_ids = processor.get_decoder_prompt_ids(language="az", task="transcribe")

with torch.no_grad():
    ids = model.generate(inputs.input_features, forced_decoder_ids=forced_ids)

text = processor.batch_decode(ids, skip_special_tokens=True)[0]
print(text)
```

> **Note:** Audio must be 16kHz mono. If your audio has a different sample rate, use `librosa.resample()` as shown above. Passing audio without resampling will produce incorrect results.

## Requirements

```bash
pip install transformers torch soundfile librosa
```


## Benchmark Details

All models evaluated on [FLEURS Azerbaijani](https://huggingface.co/datasets/LocalDoc/fleurs-azerbaijani-asr) test split (921 samples) with the same normalization (lowercase, no punctuation).

| Model | Params | WER | CER | RTF (GPU) |
|-------|--------|-----|-----|-----------|
| whisper-tiny | 38M | 104.48% | 53.93% | 0.033 |
| whisper-base | 73M | 82.63% | 30.35% | 0.032 |
| whisper-small | 242M | 52.17% | 14.52% | 0.053 |
| whisper-medium | 769M | 34.54% | 9.00% | 0.097 |
| whisper-large-v3 | 1543M | 21.00% | 5.51% | 0.129 |
| whisper-large-v3-turbo | 809M | 22.99% | 6.55% | 0.024 |
| **azerbaijani-whisper-small** | **242M** | **20.54%** | **5.72%** | **~0.05** |


## License

Apache 2.0