language: ja
datasets:

• common_voice
  metrics:
• wer
• cer
  tags:
• audio
• automatic-speech-recognition
• speech
• xlsr-fine-tuning-week
  license: apache-2.0
  model-index:
• name: XLSR Wav2Vec2 Japanese by Jonatas Grosman
  results:
  • task:
    name: 语音识别
    type: automatic-speech-recognition
    dataset:
    name: Common Voice ja
    type: common_voice
    args: ja
    metrics:
    • name: 测试WER
      type: wer
      value: 81.80
    • name: 测试CER
      type: cer
      value: 20.16

针对日语语音识别优化的XLSR-53大模型

本模型基于facebook/wav2vec2-large-xlsr-53，使用Common Voice 6.1、CSS10和JSUT数据集的训练集和验证集对日语进行了微调。使用时请确保语音输入为16kHz采样率。

本模型的微调得益于OVHcloud慷慨提供的GPU算力支持 :)

训练脚本详见: https://github.com/jonatasgrosman/wav2vec2-sprint

使用方式

可直接使用本模型（无需语言模型）进行语音识别...

使用HuggingSound库:

from huggingsound import SpeechRecognitionModel  

model = SpeechRecognitionModel("jonatasgrosman/wav2vec2-large-xlsr-53-japanese")  
audio_paths = ["/path/to/file.mp3", "/path/to/another_file.wav"]  

transcriptions = model.transcribe(audio_paths)  

自定义推理脚本:

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

LANG_ID = "ja"  
MODEL_ID = "jonatasgrosman/wav2vec2-large-xlsr-53-japanese"  
SAMPLES = 10  

test_dataset = load_dataset("common_voice", LANG_ID, split=f"test[:{SAMPLES}]")  

processor = Wav2Vec2Processor.from_pretrained(MODEL_ID)  
model = Wav2Vec2ForCTC.from_pretrained(MODEL_ID)  

# 数据预处理  
def speech_file_to_array_fn(batch):  
    speech_array, sampling_rate = librosa.load(batch["path"], sr=16_000)  
    batch["speech"] = speech_array  
    batch["sentence"] = batch["sentence"].upper()  
    return batch  

test_dataset = test_dataset.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, attention_mask=inputs.attention_mask).logits  

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

for i, predicted_sentence in enumerate(predicted_sentences):  
    print("-" * 100)  
    print("参考文本:", test_dataset[i]["sentence"])  
    print("识别结果:", predicted_sentence)  

评估

可在Common Voice日语测试集上按如下方式评估模型:

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

LANG_ID = "ja"  
MODEL_ID = "jonatasgrosman/wav2vec2-large-xlsr-53-japanese"  
DEVICE = "cuda"  

# 定义需忽略的字符集  
CHARS_TO_IGNORE = [...]  

wer = load_metric("wer.py")  
cer = load_metric("cer.py")  

chars_to_ignore_regex = f"[{re.escape(''.join(CHARS_TO_IGNORE))}]"  

# 加载处理器与模型  
processor = Wav2Vec2Processor.from_pretrained(MODEL_ID)  
model = Wav2Vec2ForCTC.from_pretrained(MODEL_ID)  
model.to(DEVICE)  

# 数据预处理函数  
def speech_file_to_array_fn(batch):  
    with warnings.catch_warnings():  
        warnings.simplefilter("ignore")  
        speech_array, sampling_rate = librosa.load(batch["path"], sr=16_000)  
    batch["speech"] = speech_array  
    batch["sentence"] = re.sub(chars_to_ignore_regex, "", batch["sentence"]).upper()  
    return batch  

# 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(DEVICE), attention_mask=inputs.attention_mask.to(DEVICE)).logits  
    pred_ids = torch.argmax(logits, dim=-1)  
    batch["pred_strings"] = processor.batch_decode(pred_ids)  
    return batch  

# 执行评估  
result = test_dataset.map(evaluate, batched=True, batch_size=8)  
predictions = [x.upper() for x in result["pred_strings"]]  
references = [x.upper() for x in result["sentence"]]  

print(f"WER: {wer.compute(predictions=predictions, references=references, chunk_size=1000) * 100}")  
print(f"CER: {cer.compute(predictions=predictions, references=references, chunk_size=1000) * 100}")  

测试结果:

下表展示了本模型的词错误率(WER)和字错误率(CER)。评估于2021-05-10完成，注意不同评估脚本可能导致结果差异。

引用

若需引用本模型，请使用:

@misc{grosman2021xlsr53-large-japanese,  
  title={针对日语语音识别优化的{XLSR}-53大模型},  
  author={Grosman, Jonatas},  
  howpublished={\url{https://huggingface.co/jonatasgrosman/wav2vec2-large-xlsr-53-japanese}},  
  year={2021}  
}