Wav2vec2 Large Xlsr Upper Sorbian Mixed

🚀 Wav2Vec2-Large-XLSR-Upper-Sorbian

本项目是在Upper Sorbian Common Voice数据集上对 facebook/wav2vec2-large-xlsr-53 进行微调得到的模型，此外还加入了来自在线 Sorbian课程 的28分钟额外音频。

使用此模型时，请确保输入的语音采样率为16kHz。

🚀 快速开始

本模型是在Upper Sorbian Common Voice数据集上对预训练模型进行微调得到的，可用于上索布语的语音识别任务。使用时需注意语音输入的采样率。

✨ 主要特性

• 基于预训练模型 facebook/wav2vec2-large-xlsr-53 进行微调。
• 结合了 Upper Sorbian Common Voice数据集 和在线课程的额外音频数据。
• 可用于上索布语的自动语音识别任务。

📦 安装指南

文档未提及具体安装步骤，故跳过此章节。

💻 使用示例

基础用法

import torch
import torchaudio
from datasets import load_dataset
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
test_dataset = load_dataset("common_voice", "hsb", split="test[:2%]")
processor = Wav2Vec2Processor.from_pretrained("jimregan/wav2vec2-large-xlsr-upper-sorbian-mixed")
model = Wav2Vec2ForCTC.from_pretrained("jimregan/wav2vec2-large-xlsr-upper-sorbian-mixed")
resampler = torchaudio.transforms.Resample(48_000, 16_000)
# Preprocessing the datasets.
# We need to read the audio 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)
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])

高级用法

import torch
import torchaudio
from datasets import load_dataset, load_metric
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
import re
test_dataset = load_dataset("common_voice", "ga-IE", split="test")
wer = load_metric("wer")
processor = Wav2Vec2Processor.from_pretrained("jimregan/wav2vec2-large-xlsr-upper-sorbian-mixed")
model = Wav2Vec2ForCTC.from_pretrained("jimregan/wav2vec2-large-xlsr-upper-sorbian-mixed")
model.to("cuda")
chars_to_ignore_regex = '[\\\\\\\\\\\\\\\\,\\\\\\\\\\\\\\\\?\\\\\\\\\\\\\\\\.\\\\\\\\\\\\\\\\!\\\\\\\\\\\\\\\\-\\\\\\\\\\\\\\\\;\\\\\\\\\\\\\\\\:\\\\\\\\\\\\\\\\"\\\\\\\\\\\\\\\\“\\\\\\\\\\\\\\\\%\\\\\\\\\\\\\\\\‘\\\\\\\\\\\\\\\\”\\\\\\\\\\\\\\\\�„«»–]'
resampler = torchaudio.transforms.Resample(48_000, 16_000)
# Preprocessing the datasets.
# We need to read the audio files as arrays
def speech_file_to_array_fn(batch):
    batch["sentence"] = remove_special_characters(batch["sentence"])
    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)
# Preprocessing the datasets.
# We need to read the audio 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"])))

📚 详细文档

测试结果

测试结果：48.2 %

训练信息

• 训练数据：使用了Common Voice的 train 和 validation 数据集，并结合了在线 Sorbian课程 中英语A1课程的词汇。
• 训练脚本：可在 此处 找到。
• 数据清理脚本：用于清理词汇数据转录的脚本在 此处。

📄 许可证

本项目采用 apache-2.0 许可证。

模型信息表格