语言: 雅库特语(sah) 数据集:

• 通用语音(common_voice) 评估指标:
• 词错误率(wer) 标签:
• 音频
• 自动语音识别
• 语音
• XLSR微调周 许可证: Apache-2.0 模型索引:
• 名称: Anton Lozhkov训练的雅库特语XLSR Wav2Vec2大模型53版 结果:
  • 任务: 名称: 语音识别 类型: 自动语音识别 数据集: 名称: 通用语音雅库特语 类型: common_voice 参数: sah 评估指标:
    • 名称: 测试词错误率 类型: wer 数值: 32.23

Wav2Vec2大模型XLSR-53雅库特语版

基于facebook/wav2vec2-large-xlsr-53模型，使用通用语音数据集对雅库特语进行微调。使用本模型时，请确保语音输入采样率为16kHz。

使用方法

该模型可直接使用（无需语言模型），示例如下：

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

test_dataset = load_dataset("common_voice", "sah", split="test[:2%]")

processor = Wav2Vec2Processor.from_pretrained("anton-l/wav2vec2-large-xlsr-53-sakha")
model = Wav2Vec2ForCTC.from_pretrained("anton-l/wav2vec2-large-xlsr-53-sakha")

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

# 数据集预处理
# 需要将音频文件读取为数组
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("预测结果:", processor.batch_decode(predicted_ids))
print("参考文本:", test_dataset["sentence"][:2])

评估

可通过以下方式在通用语音的雅库特语测试数据上评估模型：

import torch
import torchaudio
import urllib.request
import tarfile
import pandas as pd
from tqdm.auto import tqdm
from datasets import load_metric
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor

# 为节省磁盘空间，直接下载原始数据而非使用HF数据集
data_url = "https://voice-prod-bundler-ee1969a6ce8178826482b88e843c335139bd3fb4.s3.amazonaws.com/cv-corpus-6.1-2020-12-11/sah.tar.gz"
filestream = urllib.request.urlopen(data_url)
data_file = tarfile.open(fileobj=filestream, mode="r|gz")
data_file.extractall()

wer = load_metric("wer")

processor = Wav2Vec2Processor.from_pretrained("anton-l/wav2vec2-large-xlsr-53-sakha")
model = Wav2Vec2ForCTC.from_pretrained("anton-l/wav2vec2-large-xlsr-53-sakha")
model.to("cuda")

cv_test = pd.read_csv("cv-corpus-6.1-2020-12-11/sah/test.tsv", sep='\t')
clips_path = "cv-corpus-6.1-2020-12-11/sah/clips/"

def clean_sentence(sent):
    sent = sent.lower()
    # 将非字母字符替换为空格
    sent = "".join(ch if ch.isalpha() else " " for ch in sent)
    # 去除重复空格
    sent = " ".join(sent.split())
    return sent

targets = []
preds = []

for i, row in tqdm(cv_test.iterrows(), total=cv_test.shape[0]):
    row["sentence"] = clean_sentence(row["sentence"])
    speech_array, sampling_rate = torchaudio.load(clips_path + row["path"])
    resampler = torchaudio.transforms.Resample(sampling_rate, 16_000)
    row["speech"] = resampler(speech_array).squeeze().numpy()

    inputs = processor(row["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)

    targets.append(row["sentence"])
    preds.append(processor.batch_decode(pred_ids)[0])

print("词错误率: {:2f}".format(100 * wer.compute(predictions=preds, references=targets)))

测试结果: 32.23%

训练过程

训练使用了通用语音数据集的train和validation子集。