许可协议: cc-by-nc-4.0
支持语言:

• 客家话
  任务类型: 自动语音识别
  基础模型:
• openai/whisper-large-v3
  库名称: transformers

whisper-large-v3-taiwanese-hakka 模型卡片

本模型是基于台湾客家话版本openai/whisper-large-v3的微调版本，在训练过程中使用各方言ID作为提示词，旨在验证多方言场景下添加提示词是否能提升Whisper模型的微调效果。

方言与对应ID

• 四县腔: htia_sixian
• 海陆腔: htia_hailu
• 大埔腔: htia_dapu
• 饶平腔: htia_raoping
• 诏安腔: htia_zhaoan
• 南四县腔: htia_nansixian

训练过程

模型训练采用以下超参数配置：

• 批处理大小: 32
• 训练轮次: 3
• 预热步数: 50
• 总训练步数: 42549
• 学习率: 7e-5
• 数据增强: 未使用

使用方式

import torch
from transformers import AutoModelForSpeechSeq2Seq, AutoProcessor, pipeline
device = "cuda:0" if torch.cuda.is_available() else "cpu"
torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32
model_id = "formospeech/whisper-large-v3-taiwanese-hakka"
dialect_id = "htia_sixian"
model = AutoModelForSpeechSeq2Seq.from_pretrained(
    model_id, torch_dtype=torch_dtype, low_cpu_mem_usage=True, use_safetensors=True
)
model.to(device)
processor = AutoProcessor.from_pretrained(model_id)
pipe = pipeline(
    "automatic-speech-recognition",
    model=model,
    tokenizer=processor.tokenizer,
    feature_extractor=processor.feature_extractor,
    max_new_tokens=128,
    chunk_length_s=30,
    batch_size=16,
    torch_dtype=torch_dtype,
    device=device,
)
generate_kwargs = {"language": "Chinese", "prompt_ids": torch.from_numpy(processor.get_prompt_ids(dialect_id)).to(device)}
transcription = pipe("path/to/my_audio.wav", generate_kwargs=generate_kwargs)
print(transcription.replace(f" {dialect_id}", ""))