Update app.py

app.py

@@ -10,13 +10,17 @@ import openai
 from llama_index import GPTVectorStoreIndex, SimpleDirectoryReader, Prompt
 from transformers import pipeline
 import opencc
+import scipy
+import torch
 
 converter = opencc.OpenCC('t2s')  # 创建一个OpenCC实例，指定繁体字转为简体字
 ocr = CnOcr()  # 初始化ocr模型
 history_max_len = 500  # 机器人记忆的最大长度
 all_max_len = 2000  # 输入的最大长度
 asr_model_id = "openai/whisper-tiny"  # 更新为你的模型ID
-asr_pipe = pipeline("automatic-speech-recognition", model=asr_model_id)
+device = "cuda:0" if torch.cuda.is_available() else "cpu"
+asr_pipe = pipeline("automatic-speech-recognition", model=asr_model_id, device=device)
+synthesiser = pipeline("text-to-speech", "suno/bark-small", device=device)
 
 
 def get_text_emb(open_ai_key, text):  # 文本向量化
@@ -140,11 +144,23 @@ def get_response_by_llama_index(open_ai_key, msg, bot, query_engine):  # 获取
     return bot[max(0, len(bot) - 3):]  # 返回最近3轮的历史记录
 
 
+import hashlib
+
+
+def get_audio_answer(answer):  # 获取语音回答
+    speech = synthesiser(answer, forward_params={"do_sample": True})  # 生成语音
+    md5 = hashlib.md5(answer.encode('utf-8')).hexdigest()  # 获取md5
+    scipy.io.wavfile.write("{}.wav".format(md5), rate=speech["sampling_rate"], data=speech["audio"]) # 保存语音
+    return "{}.wav".format(md5)
+
+
 def get_response(open_ai_key, msg, bot, doc_text_list, doc_embeddings, query_engine, index_type):  # 获取机器人回复
     if index_type == 1:  # 如果是使用自己的索引
-        return get_response_by_self(open_ai_key, msg, bot, doc_text_list, doc_embeddings)
+        bot = get_response_by_self(open_ai_key, msg, bot, doc_text_list, doc_embeddings)
     else:  # 如果是使用llama_index索引
-        return get_response_by_llama_index(open_ai_key, msg, bot, query_engine)
+        bot = get_response_by_llama_index(open_ai_key, msg, bot, query_engine)
+    audio_answer_dir = get_audio_answer(bot[-1][1])  # 获取语音回答
+    return bot, gr.Audio(audio_answer_dir)
 
 
 def up_file(files):  # 上传文件
@@ -235,6 +251,7 @@ with gr.Blocks() as demo:
         with gr.Column():
             md = gr.Markdown("""操作说明 step 1：点击左侧区域，上传PDF，进行解析""")  # 操作说明
             chat_bot = gr.Chatbot(visible=False)  # 聊天机器人
+            audio_answer = gr.Audio()  # 语音回答
             with gr.Row():
                 asr_type = gr.Radio(value='self', choices=['self', 'openai'], label='语音识别方式', visible=False)  # 语音识别方式
                 audio_inputs = gr.Audio(source="microphone", type="filepath", label="点击录音输入", visible=False)  # 录音输入
@@ -250,7 +267,7 @@ with gr.Blocks() as demo:
     audio_inputs.change(transcribe_speech, [open_ai_key, audio_inputs, asr_type], [msg_txt])  # 录音输入
     chat_bu.click(get_response,
                   [open_ai_key, msg_txt, chat_bot, doc_text_state, doc_emb_state, query_engine, index_type],
-                  [chat_bot])  # 发送消息
+                  [chat_bot, audio_answer])  # 发送消息
 
 if __name__ == "__main__":
-    demo.queue().launch()
+    demo.queue(concurrency_count=4).launch()