đ Wav2Vec2-Large-XLSR-Indonesian
This model is a fine - tuned version of facebook/wav2vec2-large-xlsr-53 on the Indonesian Artificial Common Voice dataset. It's designed for automatic speech recognition tasks, and when using it, ensure your speech input is sampled at 16kHz.
Property |
Details |
Model Type |
XLSR Wav2Vec2 Indonesian with Artificial Voice by Cahya |
Training Data |
Artificial Common Voice train , validation , etc. datasets |
Metrics |
Word Error Rate (WER) |
Tags |
audio, automatic - speech - recognition, speech, xlsr - fine - tuning - week |
License |
apache - 2.0 |
đ Quick Start
This fine - tuned model is based on facebook/wav2vec2-large-xlsr-53 and trained on the Indonesian Artificial Common Voice dataset. Remember to sample your speech input at 16kHz when using this model.
đģ Usage Examples
Basic Usage
import torch
import torchaudio
from datasets import load_dataset
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
test_dataset = load_dataset("common_voice", "id", split="test[:2%]")
processor = Wav2Vec2Processor.from_pretrained("cahya/wav2vec2-large-xlsr-indonesian")
model = Wav2Vec2ForCTC.from_pretrained("cahya/wav2vec2-large-xlsr-indonesian")
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[:2]["speech"], 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[:2]["sentence"])
đ Documentation
Evaluation
The model can be evaluated on the Indonesian test data of Common Voice as follows:
import torch
import torchaudio
from datasets import load_dataset, load_metric
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
import re
test_dataset = load_dataset("common_voice", "id", split="test")
wer = load_metric("wer")
processor = Wav2Vec2Processor.from_pretrained("cahya/wav2vec2-large-xlsr-indonesian")
model = Wav2Vec2ForCTC.from_pretrained("cahya/wav2vec2-large-xlsr-indonesian")
model.to("cuda")
chars_to_ignore_regex = '[\,\?\.\!\-\;\:\"\â\%\â\'\â\īŋŊ]'
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"])))
Test Result: 51.69 %
Training
The Artificial Common Voice train
, validation
, and other datasets were used for training.
The script used for training can be found here
(will be available soon)
đ License
This project is licensed under the apache - 2.0 license.