Create app.py

app.py

@@ -0,0 +1,72 @@
+import torch
+from transformers import AutoModelForSpeechSeq2Seq, AutoProcessor, pipeline
+from datasets import load_dataset
+import spacy
+import gradio as gr
+
+# 設置設備
+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,
+)
+
+# DeepSeek 模型初始化（文本生成）
+deepseek_pipe = pipeline("text-generation", model="deepseek-ai/DeepSeek-R1", trust_remote_code=True)
+
+# spaCy 初始化（文本分類與標籤）
+nlp = spacy.load("en_core_web_sm")
+
+# 定義處理函數
+def process_audio(audio_file):
+    # 語音轉文字
+    result = whisper_pipe(audio_file)["text"]
+    
+    # 使用 DeepSeek 生成回應
+    messages = [{"role": "user", "content": result}]
+    deepseek_response = deepseek_pipe(messages)[0]["generated_text"]
+    
+    # 使用 spaCy 分析文本
+    doc = nlp(deepseek_response)
+    entities = [(ent.text, ent.label_) for ent in doc.ents]
+    
+    return result, deepseek_response, entities
+
+# Gradio 界面設計
+def interface(audio_file):
+    transcription, response, entities = process_audio(audio_file)
+    return {
+        "Transcription (Whisper)": transcription,
+        "AI Response (DeepSeek)": response,
+        "Extracted Entities (spaCy)": entities,
+    }
+
+# Gradio 應用程序
+with gr.Blocks() as app:
+    gr.Markdown("# AI 客服自動化系統")
+    
+    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=interface, inputs=audio_input, outputs=output_text)
+
+# 啟動應用程序（本地測試時使用）
+if __name__ == "__main__":
+    app.launch()
+
+# 部署到 Hugging Face Spaces 時，將 `app.launch()` 替換為 `app`