---
license: apache-2.0
language:
- uz
- en
- ru
base_model:
- openai/whisper-small
pipeline_tag: automatic-speech-recognition
tags:
- automatic-speech-recognition
- whisper
- speech-to-text
- uzbek
- multilingual
- transformers
- safetensors
library_name: transformers
---

# Uzbek Whisper Small STT

This repository contains a Whisper Small speech-to-text model prepared for
Hugging Face upload. The model is based on `openai/whisper-small` and is
configured to transcribe Uzbek by default.

The model files are ready to use with the `transformers` automatic speech
recognition pipeline.

## Model Details

- Model type: Whisper Small / Automatic Speech Recognition
- Default task: transcription
- Default language: Uzbek (`uz`)
- Supported languages in tokenizer/config: Uzbek, English, Russian, and other
  Whisper multilingual tokens
- Base model: `openai/whisper-small`
- Format: `safetensors`
- License: Apache-2.0

## Usage

Replace `YOUR_USERNAME/uz-whisper-small-stt` with your Hugging Face repository
name after upload.

```python
from transformers import pipeline

asr = pipeline(
    "automatic-speech-recognition",
    model="YOUR_USERNAME/uz-whisper-small-stt",
)

result = asr("audio.wav")
print(result["text"])
```

For direct generation:

```python
import torch
import torchaudio
from transformers import WhisperForConditionalGeneration, WhisperProcessor

model_id = "YOUR_USERNAME/uz-whisper-small-stt"

processor = WhisperProcessor.from_pretrained(model_id)
model = WhisperForConditionalGeneration.from_pretrained(model_id)

audio, sr = torchaudio.load("audio.wav")
if audio.shape[0] > 1:
    audio = audio.mean(dim=0, keepdim=True)

inputs = processor(
    audio.squeeze().numpy(),
    sampling_rate=sr,
    return_tensors="pt",
)

with torch.no_grad():
    predicted_ids = model.generate(inputs.input_features)

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

## Evaluation

No verified benchmark results are included in this prepared upload package.
Add WER/CER numbers here only after testing on your own evaluation set.

## Upload

Install the Hugging Face CLI and login:

```bash
pip install -U "huggingface_hub[cli]"
huggingface-cli login
```

Create a repository on Hugging Face, then upload this folder:

```bash
huggingface-cli upload YOUR_USERNAME/uz-whisper-small-stt . --repo-type model
```

Run the upload command from inside this folder:

```bash
cd hf_ready_model
```