license: apache-2.0 language: de library_name: transformers thumbnail: null tags:

• 自动语音识别
• whisper-event datasets:
• mozilla-foundation/common_voice_11_0 metrics:
• 词错误率 model-index:
• name: 针对德语ASR优化的whisper-small模型 results:
  • task: name: 自动语音识别 type: 自动语音识别 dataset: name: Common Voice 11.0 type: mozilla-foundation/common_voice_11_0 config: de split: 测试集 args: de metrics:
    • name: 贪婪解码词错误率 type: 词错误率 value: 11.35

针对德语ASR优化的whisper-small模型

本模型是基于openai/whisper-small在mozilla-foundation/common_voice_11_0德语数据集上微调的版本。使用时请确保语音输入采样率为16kHz。该模型可预测大小写和标点符号。

性能表现

下表为预训练模型在Common Voice 9.0的词错误率(WER)，结果引自原始论文。

下表为微调模型在Common Voice 11.0的表现。

使用方式

使用🤗 Pipeline推理

import torch

from datasets import load_dataset
from transformers import pipeline

device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")

# 加载pipeline
pipe = pipeline("automatic-speech-recognition", model="bofenghuang/whisper-small-cv11-german", device=device)

# 注意：为生成工具设置强制解码器ID
pipe.model.config.forced_decoder_ids = pipe.tokenizer.get_decoder_prompt_ids(language="de", task="transcribe")

# 加载数据
ds_mcv_test = load_dataset("mozilla-foundation/common_voice_11_0", "de", split="test", streaming=True)
test_segment = next(iter(ds_mcv_test))
waveform = test_segment["audio"]

# 注意：解码选项
# 限制生成token最大数为225
pipe.model.config.max_length = 225 + 1
# 采样解码
# pipe.model.config.do_sample = True
# 束搜索
# pipe.model.config.num_beams = 5
# 返回设置
# pipe.model.config.return_dict_in_generate = True
# pipe.model.config.output_scores = True
# pipe.model.config.num_return_sequences = 5

# 执行
generated_sentences = pipe(waveform)["text"]

使用🤗底层API推理

import torch
import torchaudio

from datasets import load_dataset
from transformers import AutoProcessor, AutoModelForSpeechSeq2Seq

device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")

# 加载模型
model = AutoModelForSpeechSeq2Seq.from_pretrained("bofenghuang/whisper-small-cv11-german").to(device)
processor = AutoProcessor.from_pretrained("bofenghuang/whisper-small-cv11-german", language="german", task="transcribe")

# 注意：为生成工具设置强制解码器ID
model.config.forced_decoder_ids = processor.get_decoder_prompt_ids(language="de", task="transcribe")

# 16kHz采样率
model_sample_rate = processor.feature_extractor.sampling_rate

# 加载数据
ds_mcv_test = load_dataset("mozilla-foundation/common_voice_11_0", "de", split="test", streaming=True)
test_segment = next(iter(ds_mcv_test))
waveform = torch.from_numpy(test_segment["audio"]["array"])
sample_rate = test_segment["audio"]["sampling_rate"]

# 重采样
if sample_rate != model_sample_rate:
    resampler = torchaudio.transforms.Resample(sample_rate, model_sample_rate)
    waveform = resampler(waveform)

# 提取特征
inputs = processor(waveform, sampling_rate=model_sample_rate, return_tensors="pt")
input_features = inputs.input_features
input_features = input_features.to(device)

# 生成
generated_ids = model.generate(inputs=input_features, max_new_tokens=225)  # 贪婪解码
# generated_ids = model.generate(inputs=input_features, max_new_tokens=225, num_beams=5)  # 束搜索

# 解码文本
generated_sentences = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]

# 如有需要可对预测句子进行标准化处理