基于streamlit的大模型语音交互代码和前端代码展示

streamlit介绍

Streamlit 是一款专为数据科学家和机器学习工程师打造的开源 Python 库，具有上手快、效率高的特点，它提供了丰富的功能，包括但不限于：按钮、滑块、图表等多样化组件助力打造交互界面；能读取多种格式数据并展示、支持实时更新，方便数据处理与可视化；可集成大模型、机器学习库，轻松构建模型推理、大模型交互应用；还支持一键本地部署及云部署分享，凭借热重载等特性提升开发体验，广泛适用于数据应用开发、模型演示、大模型交互等场景。

 详情见：st.chat_message - Streamlit Docs

效果展示

本次使用了llama模型，llama模型的API_KEY申请方法见：主流AI大模型的python代码调用和代码示例-CSDN博客

基于文本向量库的构建可见：基于PDF文献的本地向量库构建和大模型检索（LLM-向量数据库）-CSDN博客 

代码展示：

1.导包需求：

import docx
import time
import streamlit as st
from openai import OpenAI
from PyPDF2 import PdfReader
from langchain_core.documents import Document
import speech_recognition as sr
import pyttsx3

2. 初始化语音引擎和语音识别

# 语音引擎初始化
engine = pyttsx3.init()

def speak(text):
    """将文本转换为语音输出"""
    try:
        engine.say(text)
        engine.runAndWait()
    except Exception as e:
        st.error(f"语音播放失败: {e}")
        st.session_state.audio_error = str(e)  # 保存错误信息到会话状态

def recognize_speech():
    """从麦克风录制音频并转换为文本"""
    recognizer = sr.Recognizer()
    from speech_recognition import recognizers
    try:
        with sr.Microphone() as source:
            st.session_state.recording = True
            st.write("正在检测麦克风...")
            recognizer.adjust_for_ambient_noise(source, duration=1)

            audio = None
            st.write("请开始说话...")
            audio = recognizer.listen(source, timeout=5, phrase_time_limit=10)
            st.write("录音完成，正在识别...")

        text = recognizer.recognize_sphinx(audio, language='zh-CN')
        #text = text.hypothesis()
        return text
    except sr.WaitTimeoutError:
        return "等待超时，请重试"
    except sr.UnknownValueError:
        return "无法识别语音内容"
    except sr.RequestError as e:
        return f"服务不可用：{str(e)}"
    except Exception as e:
        st.error(f"录音错误：{str(e)}")
        return f"录音失败：{str(e)}"
    finally:
        st.session_state.recording = False

3. 定义文件处理函数

def get_pdf_text(pdf_docs):

    docs = []
    for document in pdf_docs:
        if document.type == "application/pdf":
            pdf_reader = PdfReader(document)
            for idx, page in enumerate(pdf_reader.pages):
                docs.append(
                    Document(
                        page_content=page.extract_text(),
                        metadata={"source": f"{document.name} on page {idx}"},
                    )
                )
        elif (
            document.type
            == "application/vnd.openxmlformats-officedocument.wordprocessingml.document"
        ):
            doc = docx.Document(document)
            for idx, paragraph in enumerate(doc.paragraphs):
                docs.append(
                    Document(
                        page_content=paragraph.text,
                        metadata={"source": f"{document.name} in paragraph {idx}"},
                    )
                )
    return docs

4.定义大模型交互函数

def chat_stream(query, system_message=None, temperature=None, option=None, docs=None):
    if system_message:
        messagesHistory.append({"role": "system", "content": system_message})
    #提示词模板
    prompt_template1 = (
        f"根据上传的文件内容\n{docs}\n 和问题\n{user_query}\n分析电池的性能，最后生成文献亮点总结的分析报告。"
    )
    messagesHistory.append({"role": "user", "content": prompt_template1})
    client = OpenAI(
        base_url="https://integrate.api.nvidia.com/v1",
        api_key="API_KEY"
    )
    response = client.chat.completions.create(
        model=option,
        messages=messagesHistory,
        stream=True,
        temperature=temperature
    )
    return response

5.streamlit交互系统

#信息容器初始化
if "messagesHistory" not in st.session_state:
    messagesHistory = []

if "docs" not in st.session_state:
    st.session_state.docs = None

#设置界面标题
st.title("电池交互LLM")
st.title('hello')
st.write(
    """
    您好！
    """
)
# 初始化聊天记录
if "messages" not in st.session_state:
    st.session_state.messages = [{
        "role": "assistant",
        "content": "Hi,我是BatteryChatBot~ 很高兴见到你"
    }]

# 从app历史回复中显示聊天信息
for message in st.session_state.messages:
    with st.chat_message(message["role"]):
        st.markdown(message["content"])


with st.sidebar:
    st.markdown(f"""
    <center>
    <h1>电池专家</h1>
    <h1>BatteryChatBot <sup>🤖</sup><h1/>
    </center>
    """, unsafe_allow_html=True)

    # 角色定义输入框 page_message
    system_message = st.text_area("专业领域", "您是一个电池领域的AI助手")
    # 模型选择
    company = {
        'llama': ['meta/llama-3.3-70b-instruct'],
        'Deepseek': ['DeepSeek-R1-Distill-Llama-70B', 'Deepseek-R1'],
        '智谱AI': ['glm-4', 'glm-3', 'glm-4-9b', 'glm-4-Long', 'glm-4-plus', 'glm-4-9b:605664196::3or9hkmu'],
        'OpenAI': ['gpt-3.5-turbo', 'gpt-4', 'gpt-4o'],
        '百度': ['ERNIE 4.0', 'ERNIE 4.0 Turbo', 'ERNIE 3.5'],
    }
    company_name = st.selectbox(
        label='请选择调用的模型公司',
        options=list(company.keys())
    )
    option = st.selectbox(
        label=f'{company_name} 可用模型',
        options=company[company_name]
    )
    # 文件上传控件
    pdf_docs = st.file_uploader(
        "Choose a file(上传文件)",
        # 支持的文件类型
        type=["pdf", "txt", "doc", "docx", 'json', 'xlsx'],
        # 支持多文件上传
        accept_multiple_files=True,
    )
    if pdf_docs is not None:
        if st.button(
                "处理文档",
                on_click=lambda: setattr(st.session_state, "last_action", "pdf"),
                use_container_width=True,
        ):
            if pdf_docs:
                with st.spinner("文档处理中..."):
                    st.session_state.docs = get_pdf_text(pdf_docs)
                st.success("文档处理完成，可以使用了")

    temperature = st.slider("temperature", min_value=0.0, max_value=2.0, value=1.0, step=0.1, help="值越大越具有创造力",
                            format="%.1f")

# 初始化语音相关状态
if 'recording' not in st.session_state:
    st.session_state.recording = False
if 'audio_error' not in st.session_state:
    st.session_state.audio_error = None

# 语音控制按钮列布局
col1, col2 = st.columns(2)
with col1:
    if st.button("🎤 开始录音", disabled=st.session_state.recording):
        st.session_state.mic_button = True
        st.session_state.user_query = None  # 清除之前的查询

with col2:
    if st.button("⏹️ 停止录音"):
        st.session_state.mic_button = False

# 显示语音识别状态
if st.session_state.recording:
    st.warning("正在录音中...")

# 显示语音错误信息
if st.session_state.audio_error:
    st.error(f"音频错误：{st.session_state.audio_error}")
    st.session_state.audio_error = None  # 显示后清除错误

# 语音输入处理逻辑
if st.session_state.get("mic_button", False):
    user_query1 = recognize_speech()
else:
    user_query1 = st.chat_input("请输入您的问题...")

user_query=user_query1

if user_query:
    # 显示用户输入的内容到聊天窗口
    with st.chat_message("user"):
        st.write(user_query)
    st.session_state.messages.append({"role": "user", "content": user_query})

    # 使用 OpenAI 服务处理用户问题，stream 流式输出 LLM 的答案
    with st.chat_message("assistant"):
        with st.spinner("🤖思考中..."):
            try:
                # 记录开始时间
                start_time = time.time()

                # 用 OpenAI 对话服务
                response = chat_stream(user_query, system_message, temperature, option, st.session_state.docs)

                # 创建显示消息的容器
                message_placeholder = st.empty()

                # AI 的答案
                ai_response = ""
                # 暂停和继续按钮
                pause_button = st.button("暂停回答", key="pause_button")

                # 增加中断功能
                is_interrupted = False

                for chunk in response:
                    if chunk.choices and chunk.choices[0].delta.content:
                        ai_response += chunk.choices[0].delta.content

                    message_placeholder.markdown(ai_response + "▌")

                    # 检查是否暂停
                    if pause_button:
                        is_interrupted = True
                        break

                    # 短暂延迟以允许 UI 更新
                    time.sleep(0.1)

                else:
                    elapsed_time = time.time() - start_time
                    runtime_message = f"模型运行时间：{elapsed_time:.2f}秒"
                    ai_response = f"Deepseek:R1结果：\n\n{ai_response}\n\n运行时间：{runtime_message}"
                    message_placeholder.markdown(ai_response)

                # 将助手的回复加入会话历史
                st.session_state.messages.append({"role": "assistant", "content": ai_response})

                # 语音播报回答
                if st.session_state.get("display_play_audio_button", False):
                    speak(ai_response)

            except Exception as e:
                st.error(f"AI 服务调用失败: {e}")

st.write("欢迎使用BatteryChatBot!")