语言: 波兰语
许可证: Apache-2.0
标签:

• 音频
• 自动语音识别
• 语音
• XLSR微调周
  数据集:
• Common Voice
  评估指标:
• 词错误率 (WER)
  基础模型: facebook/wav2vec2-large-xlsr-53
  模型索引:
• 名称: mbien/wav2vec2-large-xlsr-polish
  结果:
  • 任务:
    类型: 自动语音识别
    名称: 语音识别
    数据集:
    名称: Common Voice 波兰语
    类型: common_voice
    参数: pl
    评估指标:
    • 类型: WER
      值: 23.01
      名称: 测试集词错误率

Wav2Vec2-Large-XLSR-53-波兰语版

本模型基于facebook/wav2vec2-large-xlsr-53在Common Voice波兰语数据集上微调而成。使用时请确保语音输入采样率为16kHz。

使用方法

无需语言模型即可直接使用：

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

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

processor = Wav2Vec2Processor.from_pretrained("mbien/wav2vec2-large-xlsr-polish")
model = Wav2Vec2ForCTC.from_pretrained("mbien/wav2vec2-large-xlsr-polish")

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

# 数据预处理：将音频文件转为数组
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

在Common Voice波兰语测试集上的评估方法如下：

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

test_dataset = load_dataset("common_voice", "pl", split="test")
wer = load_metric("wer")

processor = Wav2Vec2Processor.from_pretrained("mbien/wav2vec2-large-xlsr-polish")
model = Wav2Vec2ForCTC.from_pretrained("mbien/wav2vec2-large-xlsr-polish")
model.to("cuda")

chars_to_ignore_regex = '[\—\…\,\?\.\!\-\;\:\"\“\„\%\‘\”\�\«\»\'\’]'
resampler = torchaudio.transforms.Resample(48_000, 16_000)

# 数据预处理
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)

# 评估函数
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("词错误率: {:.2f}%".format(100 * wer.compute(predictions=result["pred_strings"], references=result["sentence"])))

测试结果: 23.01%

训练

使用Common Voice的train和validation数据集进行训练。
训练脚本详见此处