语言: 上索布语
数据集:

• common_voice
  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: Adam Montgomerie的XLSR Wav2Vec2索布语模型
  结果:
  • 任务:
    名称: 语音识别
    类型: 自动语音识别
    数据集:
    名称: Common Voice上索布语
    类型: common_voice
    参数: hsb
    指标:
    • 名称: 测试WER
      类型: wer
      值: 41.74

Wav2Vec2-Large-XLSR-53-上索布语

基于facebook/wav2vec2-large-xlsr-53在Common Voice上索布语数据微调的模型。使用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）如下：

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("iarfmoose/wav2vec2-large-xlsr-sorbian")  
model = Wav2Vec2ForCTC.from_pretrained("iarfmoose/wav2vec2-large-xlsr-sorbian")  

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

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

可在Common Voice的上索布语测试数据上如下评估模型：

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

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

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

chars_to_ignore_regex = '[\\\\\\\\\\\\\\\\,\\\\\\\\\\\\\\\\?\\\\\\\\\\\\\\\\.\\\\\\\\\\\\\\\\!\\\\\\\\\\\\\\\\-\\\\\\\\\\\\\\\\;\\\\\\\\\\\\\\\\:\\\\\\\\\\\\\\\\"\\\\\\\\\\\\\\\\“\\\\\\\\\\\\\\\\%\\\\\\\\\\\\\\\\‘\\\\\\\\\\\\\\\\”\\\\\\\\\\\\\\\\�\\\\\\\\\\\\\\\\–\\\\\\\\\\\\\\\\—\\\\\\\\\\\\\\\\¬\\\\\\\\\\\\\\\\⅛]'  
resampler = torchaudio.transforms.Resample(48_000, 16_000)  

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)  

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"])))  

测试结果: 41.74%

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

评估脚本的笔记本详见此处