language:

• 葡萄牙语 license: apache-2.0 tags:
• 自动语音识别
• hf-asr-leaderboard
• mozilla-foundation/common_voice_8_0
• 葡萄牙语
• robust-speech-event datasets:
• mozilla-foundation/common_voice_8_0 model-index:
• name: Jonatas Grosman的XLS-R Wav2Vec2葡萄牙语模型 results:
  • task: name: 自动语音识别 type: automatic-speech-recognition dataset: name: Common Voice 8 type: mozilla-foundation/common_voice_8_0 args: pt metrics:
    • name: 测试集WER type: wer value: 8.7
    • name: 测试集CER type: cer value: 2.55
    • name: 测试集WER (+语言模型) type: wer value: 6.04
    • name: 测试集CER (+语言模型) type: cer value: 1.98
  • task: name: 自动语音识别 type: automatic-speech-recognition dataset: name: Robust Speech Event - 开发集 type: speech-recognition-community-v2/dev_data args: pt metrics:
    • name: 开发集WER type: wer value: 24.23
    • name: 开发集CER type: cer value: 11.3
    • name: 开发集WER (+语言模型) type: wer value: 19.41
    • name: 开发集CER (+语言模型) type: cer value: 10.19
  • task: name: 自动语音识别 type: automatic-speech-recognition dataset: name: Robust Speech Event - 测试集 type: speech-recognition-community-v2/eval_data args: pt metrics:
    • name: 测试集WER type: wer value: 18.8

针对葡萄牙语语音识别微调的XLS-R 1B模型

基于facebook/wav2vec2-xls-r-1b模型，使用Common Voice 8.0、CORAA、Multilingual TEDx和Multilingual LibriSpeech数据集的训练集和验证集对葡萄牙语进行了微调。使用本模型时，请确保语音输入采样率为16kHz。

本模型由HuggingSound工具微调完成，并感谢OVHcloud慷慨提供的GPU算力资源:)

使用方法

使用HuggingSound库：

from huggingsound import SpeechRecognitionModel

model = SpeechRecognitionModel("jonatasgrosman/wav2vec2-xls-r-1b-portuguese")
audio_paths = ["/path/to/file.mp3", "/path/to/another_file.wav"]

transcriptions = model.transcribe(audio_paths)

自行编写推理脚本：

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

LANG_ID = "pt"
MODEL_ID = "jonatasgrosman/wav2vec2-xls-r-1b-portuguese"
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
# 需要将音频文件读取为数组
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)

评估命令

1. 在mozilla-foundation/common_voice_8_0数据集的test分割上评估

python eval.py --model_id jonatasgrosman/wav2vec2-xls-r-1b-portuguese --dataset mozilla-foundation/common_voice_8_0 --config pt --split test

2. 在speech-recognition-community-v2/dev_data数据集上评估

python eval.py --model_id jonatasgrosman/wav2vec2-xls-r-1b-portuguese --dataset speech-recognition-community-v2/dev_data --config pt --split validation --chunk_length_s 5.0 --stride_length_s 1.0

引用

如需引用本模型，请使用以下格式：

@misc{grosman2021xlsr-1b-portuguese,
  title={针对葡萄牙语语音识别微调的{XLS-R} 1{B}模型},
  author={Grosman, Jonatas},
  howpublished={\url{https://huggingface.co/jonatasgrosman/wav2vec2-xls-r-1b-portuguese}},
  year={2022}
}