---
language: en
datasets:
- common_voice
- librispeech_asr
- how2
- must-c-v1
- must-c-v2
- europarl
- tedlium
tags:
- audio
- automatic-speech-recognition
license: cc-by-nc-4.0
---

# Fine-Tune Wav2Vec2 large model for English ASR


### Data for fine-tune

| Dataset      | Duration in hours |
|--------------|-------------------|
| Common Voice |              1667 |
| Europarl     |                85 |
| How2         |               356 |
| Librispeech  |               936 |
| MuST-C v1    |               407 |
| MuST-C v2    |               482 |
| Tedlium      |               482 |


### Evaluation result

| Dataset     | Duration in hours | WER w/o LM | WER with LM |
|-------------|-------------------|------------|-------------|
| Librispeech |        5.4        |     2.9    |     1.1     |
|   Tedlium   |        2.6        |     7.9    |     5.4     |


### Usage

[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/drive/1FAhtGvjRdHT4W0KeMdMMlL7sm6Hbe7dv?usp=sharing)

```python
from transformers.file_utils import cached_path, hf_bucket_url
from importlib.machinery import SourceFileLoader
from transformers import Wav2Vec2ProcessorWithLM
from IPython.lib.display import Audio
import torchaudio
import torch

# Load model & processor
model_name = "nguyenvulebinh/iwslt-asr-wav2vec-large-4500h"
model = SourceFileLoader("model", cached_path(hf_bucket_url(model_name,filename="model_handling.py"))).load_module().Wav2Vec2ForCTC.from_pretrained(model_name)
processor = Wav2Vec2ProcessorWithLM.from_pretrained(model_name)

# Load an example audio (16k)
audio, sample_rate = torchaudio.load(cached_path(hf_bucket_url(model_name, filename="tst_2010_sample.wav")))
input_data = processor.feature_extractor(audio[0], sampling_rate=16000, return_tensors='pt')

# Infer
output = model(**input_data)

# Output transcript without LM
print(processor.tokenizer.decode(output.logits.argmax(dim=-1)[0].detach().cpu().numpy()))
# and of course there's teams that have a lot more tada structures and among the best are recent graduates of kindergarten

# Output transcript with LM
print(processor.decode(output.logits.cpu().detach().numpy()[0], beam_width=100).text)
# and of course there are teams that have a lot more ta da structures and among the best are recent graduates of kindergarten
```

### Model Parameters License

The ASR model parameters are made available for non-commercial use only, under the terms of the Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0) license. You can find details at: https://creativecommons.org/licenses/by-nc/4.0/legalcode


### Contact 

nguyenvulebinh@gmail.com

[![Follow](https://img.shields.io/twitter/follow/nguyenvulebinh?style=social)](https://twitter.com/intent/follow?screen_name=nguyenvulebinh)