app.py

import torch
from transformers import AutoModelForSpeechSeq2Seq, AutoProcessor, pipeline
from datasets import load_dataset
import spacy
import gradio as gr

# 設置設備和環境變數（如有需要）
device = "cuda:0" if torch.cuda.is_available() else "cpu"
torch_dtype = torch.float16 if torch.cuda.is_available() else torch.float32

# Whisper 模型初始化（語音轉文字）
whisper_model_id = "openai/whisper-large-v3"
whisper_model = AutoModelForSpeechSeq2Seq.from_pretrained(
    whisper_model_id, torch_dtype=torch_dtype, low_cpu_mem_usage=True, use_safetensors=True)
whisper_model.to(device)
whisper_processor = AutoProcessor.from_pretrained(whisper_model_id)

whisper_pipe = pipeline(
    "automatic-speech-recognition",
    model=whisper_model,
    tokenizer=whisper_processor.tokenizer,
    feature_extractor=whisper_processor.feature_extractor,
    device=device)

# spaCy 初始化（文本分類與標籤）
nlp=None 

try:
        
            nlp=spacy.load("en_core_web_sm")
            
except Exception as e:
            
            print(f"加載spaCy模型時出現錯誤：{e}")

def process_audio(audio_file):
    
        # 語音轉文字
            
        result= whisper_pipe(audio_file)["text"]
                
        # 使用 T5 作為替代模型
        
        messages=[{"role": "user", "content": result}]
        
        deepseek_response="" 
        
        try: 
            from transformers import pipeline
            
            pipe=pipeline("text-generation",model="t5-base")
            
            deepseek_response=pipe(messages)[0]["generated_text"]
                    
                # 使用 spaCy 分析文本
                    
            doc=nlp(deepseek_response) if nlp is not None else None 
            entities=[(ent.text, ent.label_) for ent in doc.ents] if doc is not None else []
                    
            return {
                "Transcription (Whisper)": result,
                "AI Response (T5)": deepseek_response,# 修改為 T5 回應以避免與原來不同步                            
                "Extracted Entities (spaCy)": entities}
                        
        except Exception as e: 
                    return {
                        "Transcription (Whister)": result,# 保留原始轉錄內容                                
                            }
                            


with gr.Blocks() as app:

   
   
   
   with gr.Row():
       audio_input=gr.Audio(source="microphone", type="filepath", label="上傳語音")
       output_text=gr.JSON(label="結果")
       
       
       
       
       
  
  

submit_button=gr.Button("提交") 
submit_button.click(fn=lambda x: process_audio(x), inputs=[audio_input], outputs=[output_text])

if __name__ == "__main__":
     app.launch()