Wav2Vec2-Large-XLSR-53-Thai

Fine-tuned facebook/wav2vec2-large-xlsr-53 in Thai using the Common Voice When using this model, make sure that your speech input is sampled at 16kHz.

Usage

The model can be used directly (without a language model) as follows:

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)

## For Thai NLP Library, please feel free to check https://pythainlp.github.io/docs/2.2/api/tokenize.html
def th_tokenize(batch):
    batch["sentence"] = " ".join(word_tokenize(batch["sentence"], engine="newmm"))
    return batch

# Preprocessing the datasets.
# We need to read the aduio files as arrays
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])

Usage script here

Evaluation

The model can be evaluated as follows on the {language} test data of Common Voice.

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 = '[\,\?\.\!\-\;\:\"\“]'
resampler = torchaudio.transforms.Resample(48_000, 16_000)

## For Thai NLP Library, please feel free to check https://pythainlp.github.io/docs/2.2/api/tokenize.html
def th_tokenize(batch):
    batch["sentence"] = " ".join(word_tokenize(batch["sentence"], engine="newmm"))
    return batch

# Preprocessing the datasets.
# We need to read the aduio files as arrays
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)

# Preprocessing the datasets.
# We need to read the aduio files as arrays
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 % Evaluate script here

Training

The Common Voice train, validation datasets were used for training.

The script used for training can be found here

Downloads last month
108
Hosted inference API
Automatic Speech Recognition
or
This model can be loaded on the Inference API on-demand.
Evaluation results