🚀 Wav2Vec2-Large-XLSR-53-Chuvash
This model is fine-tuned from facebook/wav2vec2-large-xlsr-53 on Chuvash using the Common Voice, aiming to provide high - quality speech recognition for the Chuvash language.
Model Information
Property |
Details |
Model Type |
Wav2Vec2-Large-XLSR-53-Chuvash |
Training Data |
Common Voice (Chuvash) |
Metrics |
Word Error Rate (WER) |
License |
Apache-2.0 |
Model Performance
Task |
Dataset |
Metric |
Value |
Automatic Speech Recognition |
Common Voice cv |
Test WER |
48.40 |
⚠️ Important Note
When using this model, make sure that your speech input is sampled at 16kHz.
🚀 Quick Start
✨ Features
- Fine - tuned on Chuvash language data from Common Voice.
- Can be used for automatic speech recognition tasks without a language model.
📦 Installation
No specific installation steps are provided in the original document.
💻 Usage Examples
Basic Usage
import torch
import torchaudio
from datasets import load_dataset
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
test_dataset = load_dataset("common_voice", "cv", split="test")
processor = Wav2Vec2Processor.from_pretrained("gagan3012/wav2vec2-xlsr-chuvash")
model = Wav2Vec2ForCTC.from_pretrained("gagan3012/wav2vec2-xlsr-chuvash")
resampler = torchaudio.transforms.Resample(48_000, 16_000)
def speech_file_to_array_fn(batch):
speech_array, sampling_rate = torchaudio.load(batch["path"])
batch["speech"] = resampler(speech_array).squeeze().numpy()
return batch
test_dataset = test_dataset.map(speech_file_to_array_fn)
inputs = processor(test_dataset["speech"][:2], sampling_rate=16_000, return_tensors="pt", padding=True)
with torch.no_grad():
logits = model(inputs.input_values, attention_mask=inputs.attention_mask).logits
predicted_ids = torch.argmax(logits, dim=-1)
print("Prediction:", processor.batch_decode(predicted_ids))
print("Reference:", test_dataset["sentence"][:2])
Results:
Prediction: ['проектпа килӗшӳллӗн тӗлӗ мероприяти иртермелле', 'твăра çак планета минтӗ пуяни калленнана']
Reference: ['Проектпа килӗшӳллӗн, тӗрлӗ мероприяти ирттермелле.', 'Çак планета питĕ пуян иккен.']
Evaluation
import torch
import torchaudio
from datasets import load_dataset, load_metric
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
import re
!mkdir cer
!wget -O cer/cer.py https://huggingface.co/ctl/wav2vec2-large-xlsr-cantonese/raw/main/cer.py
test_dataset = load_dataset("common_voice", "cv", split="test")
wer = load_metric("wer")
cer = load_metric("cer")
processor = Wav2Vec2Processor.from_pretrained("gagan3012/wav2vec2-xlsr-chuvash")
model = Wav2Vec2ForCTC.from_pretrained("gagan3012/wav2vec2-xlsr-chuvash")
model.to("cuda")
chars_to_ignore_regex = '[\\\\,\\\\?\\\\.\\\\!\\\\-\\\\;\\\\:\\\\"\\\\“]'
resampler = torchaudio.transforms.Resample(48_000, 16_000)
def speech_file_to_array_fn(batch):
batch["sentence"] = re.sub(chars_to_ignore_regex, '', batch["sentence"]).lower()
speech_array, sampling_rate = torchaudio.load(batch["path"])
batch["speech"] = resampler(speech_array).squeeze().numpy()
return batch
test_dataset = test_dataset.map(speech_file_to_array_fn)
def evaluate(batch):
inputs = processor(batch["speech"], sampling_rate=16_000, return_tensors="pt", padding=True)
with torch.no_grad():
logits = model(inputs.input_values.to("cuda"), attention_mask=inputs.attention_mask.to("cuda")).logits
pred_ids = torch.argmax(logits, dim=-1)
batch["pred_strings"] = processor.batch_decode(pred_ids)
return batch
result = test_dataset.map(evaluate, batched=True, batch_size=8)
print("WER: {:2f}".format(100 * wer.compute(predictions=result["pred_strings"], references=result["sentence"])))
print("CER: {:2f}".format(100 * cer.compute(predictions=result["pred_strings"], references=result["sentence"])))
Test Result: 48.40 %
🔧 Technical Details
The script used for training can be found here
📄 License
This model is licensed under the Apache - 2.0 license.