File size: 3,252 Bytes

72f2f96

---
language:
- en
- zh
- de
- es
- ru
- ko
- fr
- ja
- pt
- tr
- pl
- ca
- nl
- ar
- sv
- it
- id
- hi
- fi
- vi
- he
- uk
- el
- ms
- cs
- ro
- da
- hu
- ta
- "no"
- th
- ur
- hr
- bg
- lt
- la
- mi
- ml
- cy
- sk
- te
- fa
- lv
- bn
- sr
- az
- sl
- kn
- et
- mk
- br
- eu
- is
- hy
- ne
- mn
- bs
- kk
- sq
- sw
- gl
- mr
- pa
- si
- km
- sn
- yo
- so
- af
- oc
- ka
- be
- tg
- sd
- gu
- am
- yi
- lo
- uz
- fo
- ht
- ps
- tk
- nn
- mt
- sa
- lb
- my
- bo
- tl
- mg
- as
- tt
- haw
- ln
- ha
- ba
- jw
- su
license: apache-2.0
tags:
- whisper
- speech-recognition
- multilingual
- automatic-speech-recognition
pipeline_tag: automatic-speech-recognition
library_name: transformers
widget:
- example_title: Librispeech sample 1
  src: https://cdn-media.huggingface.co/speech_samples/sample1.flac
- example_title: Librispeech sample 2
  src: https://cdn-media.huggingface.co/speech_samples/sample2.flac
---

# MultilingualSTT

OpenAI's Whisper Large V3 model for multilingual speech-to-text transcription.

## Model Description

Whisper Large V3 is a state-of-the-art automatic speech recognition (ASR) model that supports 99+ languages. It provides highly accurate transcription across a wide range of languages and acoustic conditions.

## Key Features

- **99+ Languages**: Supports English, Chinese, German, Spanish, Russian, Korean, French, Japanese, Portuguese, Turkish, Polish, Italian, Hindi, Arabic, and many more
- **Speech Translation**: Can translate speech to English
- **Timestamps**: Supports word-level and sentence-level timestamps
- **Robust**: Excellent handling of accents, background noise, and technical language

## Usage

```python
import torch
from transformers import AutoModelForSpeechSeq2Seq, AutoProcessor, pipeline

device = "cuda:0" if torch.cuda.is_available() else "cpu"
torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32

model_id = "Svetozar1993/MultilingualSTT"

model = AutoModelForSpeechSeq2Seq.from_pretrained(
    model_id, torch_dtype=torch_dtype, low_cpu_mem_usage=True, use_safetensors=True
)
model.to(device)

processor = AutoProcessor.from_pretrained(model_id)

pipe = pipeline(
    "automatic-speech-recognition",
    model=model,
    tokenizer=processor.tokenizer,
    feature_extractor=processor.feature_extractor,
    torch_dtype=torch_dtype,
    device=device,
)

result = pipe("audio.mp3")
print(result["text"])
```

## Advanced Usage

### Specify Language
```python
result = pipe(sample, generate_kwargs={"language": "french"})
```

### Speech Translation (to English)
```python
result = pipe(sample, generate_kwargs={"task": "translate"})
```

### Get Timestamps
```python
result = pipe(sample, return_timestamps=True)
print(result["chunks"])
```

### Word-Level Timestamps
```python
result = pipe(sample, return_timestamps="word")
```

## Flash Attention 2

For faster inference:
```bash
pip install flash-attn --no-build-isolation
```

```python
model = AutoModelForSpeechSeq2Seq.from_pretrained(
    model_id,
    torch_dtype=torch_dtype,
    low_cpu_mem_usage=True,
    attn_implementation="flash_attention_2"
)
```

## Model Details

- **Architecture:** Whisper Large V3
- **Parameters:** 1.55B
- **Languages:** 99+
- **License:** Apache 2.0

## Author

Svetozar1993