许可证：apache-2.0
标签：

• 训练生成
  指标：
• 词错误率（wer）
  模型索引：
• 名称：whisper-large-v2-spanish
  结果：[]

whisper-large-v2-spanish

该模型是基于openai/whisper-large-v2在特定数据集上微调的版本。
在评估集上取得了以下结果：

• 损失：0.1466
• 词错误率（WER）：0.0855

模型描述

需补充更多信息

预期用途与限制

需补充更多信息

训练与评估数据

需补充更多信息

训练过程

训练超参数

训练过程中使用了以下超参数：

• 学习率：1e-05
• 训练批次大小：16
• 评估批次大小：16
• 随机种子：42
• 优化器：Adam（β1=0.9，β2=0.999，ε=1e-08）
• 学习率调度器类型：线性
• 学习率预热步数：500
• 训练总步数：25000
• 混合精度训练：原生AMP

训练结果

转录示例：

from datasets import load_dataset, Audio  
import torch  
from transformers import WhisperProcessor, WhisperForConditionalGeneration  

# 设备  
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")  

# 加载模型  
processor = WhisperProcessor.from_pretrained("clu-ling/whisper-large-v2-spanish")  
model = WhisperForConditionalGeneration.from_pretrained("clu-ling/whisper-large-v2-spanish").to(device)  
forced_decoder_ids = processor.get_decoder_prompt_ids(language="es", task="transcribe")  

# 加载数据集  
commonvoice_eval = load_dataset("mozilla-foundation/common_voice_11_0", "es", split="validation", streaming=True)  
commonvoice_eval = commonvoice_eval.cast_column("audio", Audio(sampling_rate=16000))  
sample = next(iter(commonvoice_eval))["audio"]  

# 提取特征并生成标记ID  
input_features = processor(sample["array"], sampling_rate=sample["sampling_rate"], return_tensors="pt").input_features  
predicted_ids = model.generate(input_features.to(device), forced_decoder_ids=forced_decoder_ids)  

# 解码  
transcription = processor.batch_decode(predicted_ids, skip_special_tokens=True)  

print(transcription)  

评估：

在mozilla-foundation/common_voice_11_0测试集上评估该模型。

from transformers.models.whisper.english_normalizer import BasicTextNormalizer  
from datasets import load_dataset, Audio  
import evaluate  
import torch  
import re  
from transformers import WhisperProcessor, WhisperForConditionalGeneration  

# 设备  
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")  

# 评估指标  
wer_metric = evaluate.load("wer")  

# 模型  
processor = WhisperProcessor.from_pretrained("clu-ling/whisper-large-v2-spanish")  
model = WhisperForConditionalGeneration.from_pretrained("clu-ling/whisper-large-v2-spanish")  

# 数据集  
dataset = load_dataset("mozilla-foundation/common_voice_11_0", "es", split="test")  
dataset = dataset.cast_column("audio", Audio(sampling_rate=16000))  

# 调试用：仅取部分样本  
# dataset = dataset.shard(num_shards=10000, index=0)  
# print(dataset)  

def normalize(batch):  
    batch["gold_text"] = whisper_norm(batch['sentence'])  
    return batch  

def map_wer(batch):  
    model.to(device)  
    forced_decoder_ids = processor.get_decoder_prompt_ids(language="es", task="transcribe")  
    inputs = processor(batch["audio"]["array"], sampling_rate=batch["audio"]["sampling_rate"], return_tensors="pt").input_features  
    with torch.no_grad():  
        generated_ids = model.generate(inputs=inputs.to(device), forced_decoder_ids=forced_decoder_ids)  
        transcription = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]  
    batch["predicted_text"] = whisper_norm(transcription)  
    return batch  

# 处理参考文本  
processed_dataset = dataset.map(normalize)  
# 获取预测结果  
predicted = processed_dataset.map(map_wer)  

# 计算词错误率  
wer = wer_metric.compute(references=predicted['gold_text'], predictions=predicted['predicted_text'])  
wer = round(100 * wer, 2)  
print("WER:", wer)  

框架版本

• Transformers 4.26.0.dev0
• PyTorch 1.13.1+cu117
• Datasets 2.8.1.dev0
• Tokenizers 0.13.2