๐ Wav2Vec2-Large-XLSR-53-Thai
This model is a fine - tuned version of facebook/wav2vec2-large-xlsr-53 in Thai, leveraging the Common Voice dataset. Ensure your speech input is sampled at 16kHz when using this model.
๐ Quick Start
This model is a fine - tuned version of facebook/wav2vec2-large-xlsr-53 in Thai, using the Common Voice dataset. When using this model, make sure that your speech input is sampled at 16kHz.
โจ Features
- Fine - tuned for Thai language on the Common Voice dataset.
- Can be used for automatic speech recognition tasks in Thai.
๐ฆ Installation
No specific installation steps are provided in the original README. So, this section is skipped.
๐ป Usage Examples
Basic Usage
import torch
import torchaudio
from datasets import load_dataset
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
from pythainlp.tokenize import word_tokenize
test_dataset = load_dataset("common_voice", "th", split="test[:2%]")
processor = Wav2Vec2Processor.from_pretrained("sakares/wav2vec2-large-xlsr-thai-demo")
model = Wav2Vec2ForCTC.from_pretrained("sakares/wav2vec2-large-xlsr-thai-demo")
resampler = torchaudio.transforms.Resample(48_000, 16_000)
def th_tokenize(batch):
batch["sentence"] = " ".join(word_tokenize(batch["sentence"], engine="newmm"))
return batch
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).map(th_tokenize)
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])
The usage script can be found here.
๐ Documentation
Evaluation
The model can be evaluated on the Thai test data of Common Voice as follows:
import torch
import torchaudio
from datasets import load_dataset, load_metric
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
from pythainlp.tokenize import word_tokenize
import re
test_dataset = load_dataset("common_voice", "th", split="test")
wer = load_metric("wer")
processor = Wav2Vec2Processor.from_pretrained("sakares/wav2vec2-large-xlsr-thai-demo")
model = Wav2Vec2ForCTC.from_pretrained("sakares/wav2vec2-large-xlsr-thai-demo")
model.to("cuda")
chars_to_ignore_regex = '[\,\?\.\!\-\;\:\"\รขโฌล]'
resampler = torchaudio.transforms.Resample(48_000, 16_000)
def th_tokenize(batch):
batch["sentence"] = " ".join(word_tokenize(batch["sentence"], engine="newmm"))
return batch
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).map(th_tokenize)
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: 44.46 %
The evaluation script can be found here.
Training
The Common Voice train
and validation
datasets were used for training.
The training script can be found here.
๐ License
This model is licensed under the Apache 2.0 license.
๐ฆ Model Information
Property |
Details |
Model Type |
Fine - tuned Wav2Vec2 - Large - XLSR - 53 for Thai |
Training Data |
Common Voice (train and validation datasets) |
Test Dataset |
Common Voice Thai test dataset |
Test WER |
44.46% |