Wav2vec2 Base Gujarati Demo

🚀 wav2vec2-base-gujarati-demo

該項目是在古吉拉特語（Guj）上對 facebook/wav2vec2-large-xlsr-53 進行微調得到的模型，可用於語音識別任務，使用時需確保語音輸入採樣率為 16kHz。

🚀 快速開始

本模型是在古吉拉特語（Guj）上對 facebook/wav2vec2-large-xlsr-53 進行微調得到的。使用該模型時，請確保您的語音輸入採樣率為 16kHz。

✨ 主要特性

• 語言支持：支持古吉拉特語（Guj）的自動語音識別。
• 模型微調：基於 facebook/wav2vec2-large-xlsr-53 進行微調。

💻 使用示例

基礎用法

import torch
import torchaudio
from datasets import load_dataset
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor

common_voice_train,common_voice_test = load_dataset('csv', data_files={'train': 'train.csv','test': 'test.csv'},error_bad_lines=False,encoding='utf-8',split=['train', 'test'])

processor = Wav2Vec2Processor.from_pretrained("jaimin/wav2vec2-base-gujarati-demo")
model = Wav2Vec2ForCTC.from_pretrained("jaimin/wav2vec2-base-gujarati-demo")

resampler = torchaudio.transforms.Resample(48_000, 16_000)

# 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 = common_voice_test.map(speech_file_to_array_fn)
inputs = processor(test_dataset["speech"], sampling_rate=16_000, return_tensors="pt", padding=True)

with torch.no_grad():
     logits = model(inputs.input_values).logits

predicted_ids = torch.argmax(logits, dim=-1)

print("Prediction:", processor.batch_decode(predicted_ids))
print("Reference:", test_dataset["sentence"][0].lower())

高級用法

import torch
import torchaudio
from datasets import load_dataset, load_metric
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
import re

common_voice_validation = load_dataset('csv', data_files={'test': 'validation.csv'},error_bad_lines=False,encoding='utf-8',split='test')
wer = load_metric("wer")

processor = Wav2Vec2Processor.from_pretrained("jaimin/wav2vec2-base-gujarati-demo")
model = Wav2Vec2ForCTC.from_pretrained("Amrrs/wav2vec2-base-gujarati-demo")
model.to("cuda")

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

# 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 = common_voice_validation.map(speech_file_to_array_fn)

# Preprocessing the datasets.
# We need to read the aduio files as arrays
def evaluate(batch):
    inputs = processor(batch["speech"], sampling_rate=16000, return_tensors="pt", padding=True)

    with torch.no_grad():
        logits = model(inputs.input_values.to("cuda")).logits
    
    pred_ids = torch.argmax(logits, dim=-1)
    batch["pred_strings"] = processor.batch_decode(pred_ids)
    return batch

result = common_voice_validation.map(evaluate, batched=True, batch_size=8)

print("WER: {:2f}".format(100 * wer.compute(predictions=result["pred_strings"], references=result["sentence"])))

測試結果：28.92 %

📚 詳細文檔

該模型可在 Common Voice 的古吉拉特語（Guj）測試數據上進行如下評估。

🔧 技術細節

• 模型評估：使用字錯率（WER）作為評估指標。
• 訓練腳本：訓練使用的腳本可在 此處 找到。

📄 許可證

本項目採用 Apache-2.0 許可證。