Wav2vec2 Large Xlsr 53 Chinese Zn Cn Aishell1

🚀 Wav2Vec2-Large-XLSR-53-Chinese-zh-CN-aishell1

This model is fine-tuned from facebook/wav2vec2-large-xlsr-53 on Chinese using the AISHELL-1 dataset. It's designed for automatic speech recognition tasks.

🚀 Quick Start

Fine-tuned facebook/wav2vec2-large-xlsr-53 on Chinese using the AISHELL-1 dataset. When using this model, make sure that your speech input is sampled at 16kHz.

✨ Features

• Dataset: Fine-tuned on the AISHELL-1 dataset for Chinese speech recognition.
• Metrics: Evaluated using Word Error Rate (WER).
• Compatibility: Requires speech input sampled at 16kHz.

📦 Installation

No specific installation steps are provided in the original README. If you want to use this model, you need to install the necessary libraries such as torch, librosa, and transformers. You can install them using pip:

pip install torch librosa transformers

💻 Usage Examples

Basic Usage

import torch
import librosa
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor

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

processor = Wav2Vec2Processor.from_pretrained(
    'qinyue/wav2vec2-large-xlsr-53-chinese-zn-cn-aishell1')
model = Wav2Vec2ForCTC.from_pretrained(
    'qinyue/wav2vec2-large-xlsr-53-chinese-zn-cn-aishell1').to(device)

filepath = 'test.wav'
audio, sr = librosa.load(filepath, sr=16000, mono=True)
inputs = processor(audio, sample_rate=16000, return_tensors="pt").to(device)
with torch.no_grad():
    logits = model(inputs.input_values,
                   attention_mask=inputs.attention_mask).logits
predicted_ids = torch.argmax(logits, dim=-1)
pred_str = processor.decode(predicted_ids[0])

print(pred_str)

Advanced Usage

wer_metric = load_metric("wer")

def compute_metrics(pred):
    pred_logits = pred.predictions
    pred_ids = np.argmax(pred_logits, axis=-1)

    pred.label_ids[pred.label_ids == -100] = processor.tokenizer.pad_token_id

    pred_str = processor.batch_decode(pred_ids, spaces_between_special_tokens=True)
    label_str = processor.batch_decode(pred.label_ids, group_tokens=False, spaces_between_special_tokens=True)

    wer = wer_metric.compute(predictions=pred_str, references=label_str)

    return {"wer": wer}

📚 Documentation

Results

Test Result:

In the table below I report the Word Error Rate (WER) of the model on the AISHELL-1 test dataset.

📄 License

This model is licensed under the Apache-2.0 license.