tabs/chatbot_tab.py

import streamlit as st # type: ignore
import os
from datetime import datetime
from extra_streamlit_components import tab_bar, TabBarItemData
import io
from gtts import gTTS
import soundfile as sf
import wavio
from audio_recorder_streamlit import audio_recorder
import speech_recognition as sr
import whisper
import numpy as np
from translate_app import tr
import getpass
from langchain_mistralai import ChatMistralAI
from langchain_openai import ChatOpenAI
from langgraph.checkpoint.memory import MemorySaver
from langgraph.graph import START, END, MessagesState, StateGraph
from langchain_core.prompts import ChatPromptTemplate, MessagesPlaceholder
from typing import Sequence
from langchain_core.messages import BaseMessage, SystemMessage, HumanMessage, AIMessage, trim_messages
from langgraph.graph.message import add_messages
from typing_extensions import Annotated, TypedDict
from dotenv import load_dotenv
import time
from tabs.google_drive_read_preprompt import read_param, format_param
import warnings
warnings.filterwarnings('ignore')

title = "Sales coaching"
sidebar_name = "Sales coaching"
dataPath = st.session_state.DataPath

os.environ["LANGCHAIN_TRACING_V2"] = "true"
os.environ["LANGCHAIN_ENDPOINT"]="https://api.smith.langchain.com"
os.environ["LANGCHAIN_HUB_API_URL"]="https://api.smith.langchain.com"
os.environ["LANGCHAIN_PROJECT"] = "Sales Coaching Chatbot" 
if st.session_state.Cloud != 0:
    load_dotenv()
os.getenv("LANGCHAIN_API_KEY")
os.getenv("MISTRAL_API_KEY")
os.getenv("OPENAI_API_KEY")


prompt = ChatPromptTemplate.from_messages(
    [
        (
            "system",
            "Répond à toutes les questions du mieux possible dans la langue {language}, même si la question est posée dans une autre langue",
        ),
        MessagesPlaceholder(variable_name="messages"),
    ]
)

class State(TypedDict):
    messages: Annotated[Sequence[BaseMessage], add_messages]
    language: str

def call_model(state: State):
    chain = prompt | model
    response = chain.invoke(state)
    return {"messages": [response]}

# Define a new graph
workflow = StateGraph(state_schema=State)

# Define the (single) node in the graph
workflow.add_edge(START, "model")
workflow.add_node("model", call_model)
workflow.add_edge("model", END)

# Add memory
memory = MemorySaver()
app = workflow.compile(checkpointer=memory)

selected_index1 = 0
selected_index2 = 0
selected_index3 = 0
selected_indices4 = []
selected_indices5 = []    
selected_indices6 = []
selected_indices7 = [] 
selected_options4 = []
selected_options5 = []
selected_options6 = []
selected_options7 = []
selected_index8 = 0
context=""
human_message1=""
thread_id =""
virulence = 1
question = []
thread_id = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
config = {"configurable": {"thread_id": thread_id}}
to_init = True
initialized = False
messages = [
    SystemMessage(content=""),
    HumanMessage(content=""),
    AIMessage(content=""),
    HumanMessage(content="")
    ]
if 'model' in st.session_state:
    model = st.session_state.model
    used_model = st.session_state.model

def init_run():
    global initialized, to_init, thread_id, config, app, context, human_message1, model, used_model, messages

    initialized = True
    to_init = False
    thread_id = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
    config = {"configurable": {"thread_id": thread_id}}
    app.invoke(
    {"messages": messages, "language": language},
    config,
    )  
    st.session_state.thread_id = thread_id
    st.session_state.config = config
    st.session_state.messages_init = messages
    st.session_state.context = context
    st.session_state.human_message1 = human_message1
    st.session_state.messages = []
    if 'model' in st.session_state and (st.session_state.model[:3]=="gpt") and ("OPENAI_API_KEY" in st.session_state):
        model = ChatOpenAI(model=st.session_state.model,
        temperature=0.8,  # Adjust creativity level
        max_tokens=150   # Define max output token limit
        )           
    else: 
        model = ChatMistralAI(model=st.session_state.model)
    if 'model' in st.session_state:
        used_model=st.session_state.model
    return 

def init():
    global config,thread_id, context,human_message1,ai_message1,language, app, model_speech,prompt,model,question, to_init, initialized
    global selected_index1, selected_index2, selected_index3, selected_indices4,selected_indices5,selected_indices6,selected_indices7 
    global selected_options4,selected_options5,selected_options6,selected_options7, selected_index8, virulence, used_model, messages
    
    model_speech = whisper.load_model("base") 
    
    if (st.button(label=tr("Nouvelle conversation"), type="primary")):
        selected_index1 = 0
        selected_index2 = 0
        selected_index3 = 0
        selected_indices4 = []
        selected_indices5 = []    
        selected_indices6 = []
        selected_indices7 = [] 
        selected_options4 = []
        selected_options5 = []
        selected_options6 = []
        selected_options7 = []
        selected_index8 = 0
        context = ""
        human_message1=""
        thread_id =""
        virulence = 1
        if 'model' in st.session_state and (st.session_state.model[:3]=="gpt") and ("OPENAI_API_KEY" in st.session_state):
            model = ChatOpenAI(model=st.session_state.model,
            temperature=0.8,  # Adjust creativity level
            max_tokens=150   # Define max output token limit
            )           
        else: 
            model = ChatMistralAI(model=st.session_state.model)
        if 'model' in st.session_state:
            used_model=st.session_state.model
 
    label, question, options = format_param()
    translated_options1 = [tr(o) for o in options[0]]
    selected_option1 = st.selectbox(tr(label[0]),translated_options1, index = selected_index1) # index=int(var1_init))
    selected_index1 = translated_options1.index(selected_option1)
    
    translated_options2 = [tr(o) for o in options[1]]
    selected_option2 = st.selectbox(tr(label[1]),translated_options2, index = selected_index2) # index=int(var2_init))
    selected_index2 = translated_options2.index(selected_option2)
    
    translated_options3 = [tr(o) for o in options[2]]
    selected_option3 = st.selectbox(tr(label[2]),translated_options3, index=selected_index3) #index=int(var3_init))
    selected_index3 = translated_options3.index(selected_option3)

    context = tr(f"""Tu es un {options[0][selected_index1]}, d'une {options[1][selected_index2]}.
Cette entreprise propose des {options[2][selected_index3]}.
    """)
    context = st.text_area(label=tr("Résumé du Contexte (modifiable):"), value=context)
    st.markdown('''
                ------------------------------------------------------------------------------------
                ''')
    
    translated_options4 = [tr(o) for o in options[3]]
    selected_options4 = st.multiselect(tr(label[3]),translated_options4, default=[translated_options4[o] for o in selected_indices4])
    selected_indices4 = [translated_options4.index(o) for o in selected_options4]
    problematique = selected_options4
    if problematique != []:
        markdown_text4 = """\n"""+tr(question[3])
        markdown_text4 = markdown_text4+"".join(f"\n- {o}" for o in problematique)
        st.write(markdown_text4)
    else: markdown_text4 = ""
 
    translated_options5 = [tr(o) for o in options[4]]
    selected_options5 = st.multiselect(tr(label[4]),translated_options5, default=[translated_options5[o] for o in selected_indices5])
    selected_indices5 = [translated_options5.index(o) for o in selected_options5]
    processus = selected_options5
    if processus != []:
        markdown_text5 = """\n\n"""+tr(question[4])
        markdown_text5 = markdown_text5+"".join(f"\n- {o}" for o in processus)
        st.write(markdown_text5)
    else: markdown_text5 = ""
    
    translated_options6 = [tr(o) for o in options[5]]
    selected_options6 = st.multiselect(tr(label[5]),translated_options6, default=[translated_options6[o] for o in selected_indices6])
    selected_indices6 = [translated_options6.index(o) for o in selected_options6]
    objectifs = selected_options6
    if objectifs != []:
        markdown_text6 = """\n\n"""+tr(question[5])
        markdown_text6 = markdown_text6+"".join(f"\n- {o}" for o in objectifs)
        st.write(markdown_text6)
    else: markdown_text6 = ""
    
    translated_options7 = [tr(o) for o in options[6]]
    selected_options7 = st.multiselect(tr(label[6]),translated_options7, default=[translated_options7[o] for o in selected_indices7])
    selected_indices7 = [translated_options7.index(o) for o in selected_options7]
    solutions_utilisees = selected_options7
    if solutions_utilisees != []:
        markdown_text7 = """\n\n"""+tr(question[6])
        markdown_text7 = markdown_text7+"".join(f"\n- {o}" for o in solutions_utilisees)
        st.write(markdown_text7)
        st.write("")
    else: markdown_text7 = ""
    
    translated_options8 = [tr(o) for o in options[7]]
    selected_option8 = st.selectbox(tr(label[7]),translated_options8, index = selected_index8) 
    selected_index8 = translated_options8.index(selected_option8)
    markdown_text8 = """\n\n"""+tr(question[7])+"""\n"""+(f"""{translated_options8[selected_index8]}""")
    
    
    col1, col2, col3 = st.columns(3)
    with col1:
        virulence = st.slider(tr("Virulence (choisissez une valeur entre 1 et 5)"), min_value=1, max_value=5, step=1,value=virulence)
    markdown_text9 = """\n\n"""+tr(f"""Le prospect est très occupé et n'aime pas être dérangé inutilement.  
Tu vas utiliser une échelle de 1 à 5 d'agressivité du prospect à l'égard du vendeur.  
Pour cette simulation utilise le niveau {virulence}.""")
    
    human_message1 = tr("""Je souhaite que nous ayons une conversation verbale entre moi le vendeur, et toi que je prospecte.
Mon entreprise propose une solution logicielle pour gérer la proposition de valeur d’entreprise B2B qui commercialise des solutions technologiques.    
""")+markdown_text4+markdown_text5+markdown_text6+markdown_text7+markdown_text8+markdown_text9+tr(f"""
  
Je suis le vendeur.
Répond à mes questions en tant que {options[0][selected_index1]}, connaissant mal le concept de proposition de valeur,
et mon équipe de vente n'est pas performante.  
  
Attention: Ce n'est pas toi qui m'aide, c'est moi qui t'aide avec ma solution.  
Attention: Si le vendeur aborde des points qui ne concerne pas cette simulation, lui répondre que c'est hors contexte.  
  
Es tu prêt à commencer ?
""")

    human_message1 = st.text_area(label=tr("Consigne"), value=tr(human_message1),height=300)
    st.markdown('''
                ------------------------------------------------------------------------------------
                ''')
    
    ai_message1 = tr(f"J'ai bien compris, je suis un {options[0][selected_index1]} prospecté et je réponds seulement à tes questions. Je réponds à une seule question à la fois, sans commencer mes réponses par 'En tant que {options[0][selected_index1]}'.")   


    
    # ai_message1 = st.text_area(label=tr("Réponse du prospect"), value=ai_message1)
    messages = [
        SystemMessage(content=context),
        HumanMessage(content=human_message1),
        # AIMessage(content=ai_message1),
        # HumanMessage(content=tr("Commençons la conversation. Attention, je suis le vendeur et je parle le premier. Tu es le propect."))
        ]

    st.write("")
    if ("context" in st.session_state) and ("human_message1" in st.session_state):
        if (st.session_state.context != context) or (st.session_state.human_message1 != human_message1 ) or (used_model != st.session_state.model) or (thread_id==""):
            to_init = True
        else:
            to_init = False
    else:
        to_init = True
    
    if to_init:
        if st.button(label=tr("Validez"), on_click=init_run,type="primary"):
            initialized=True
        else: initialized = False
    st.write("**thread_id:** "+thread_id) 
    return config, thread_id, messages

# Fonction pour générer et jouer le texte en speech
def play_audio(custom_sentence, Lang_target, speed=1.0):
    # Générer le speech avec gTTS
    audio_stream_bytesio_src = io.BytesIO()
    tts = gTTS(custom_sentence, lang=Lang_target)
    
    # Revenir au début du flux audio
    audio_stream_bytesio_src.seek(0)
    audio_stream_bytesio_src.truncate(0)
    
    tts.write_to_fp(audio_stream_bytesio_src)

    audio_stream_bytesio_src.seek(0)
    
    # Charger l'audio dans un tableau numpy
    data, samplerate = sf.read(audio_stream_bytesio_src)

    # Modifier la vitesse de lecture en ajustant le taux d'échantillonnage
    new_samplerate = int(samplerate * speed)
    new_audio_stream_bytesio = io.BytesIO()

    # Enregistrer l'audio avec la nouvelle fréquence d'échantillonnage
    sf.write(new_audio_stream_bytesio, data, new_samplerate, format='wav')
    new_audio_stream_bytesio.seek(0)

    # Lire l'audio dans Streamlit
    # time.sleep(2)
    st.audio(new_audio_stream_bytesio, start_time=0, autoplay=True)



def run():
    global thread_id, config, model_speech, language,prompt,model, model_name, question, to_init, initialized, messages
   
    st.write("")
    st.write("")
    st.title(tr(title))
    
    if 'language_label' in st.session_state:
        language = st.session_state['language_label'] 
    else: language = "French"
    
    chosen_id = tab_bar(data=[
        TabBarItemData(id="tab1", title=tr("Initialisation"), description=tr("d'une nouvelle conversation")),
        TabBarItemData(id="tab2", title=tr("Conversation"), description=tr("avec le prospect")),
        TabBarItemData(id="tab3", title=tr("Evaluation"), description=tr("de l'acte de vente"))],          
        default="tab1")
                        
                        
    if (chosen_id == "tab1"):
        if 'model' in st.session_state and (st.session_state.model[:3]=="gpt") and ("OPENAI_API_KEY" in st.session_state):
            model = ChatOpenAI(model=st.session_state.model,
                                 temperature=0.8,  # Adjust creativity level
                                 max_tokens=150   # Define max output token limit
                                )           
        else: 
            model = ChatMistralAI(model=st.session_state.model)
            

        config,thread_id, messages = init()
        query = ""
    elif (chosen_id == "tab2"):
        try:
            if to_init and not initialized:
                init_run()
        except NameError:
            config,thread_id, messages = init()
        with st.container():
            # Diviser l'écran en deux colonnes
            col1, col2 = st.columns(2)
            with col1:
                st.write("**thread_id:** "+thread_id)
                query = ""
                audio_bytes = audio_recorder (pause_threshold=2.0,  sample_rate=16000, auto_start=False, text=tr("Cliquez pour parler, puis attendre 2sec."), \
                                            recording_color="#e8b62c", neutral_color="#1ec3bc", icon_size="6x",)
            
                if audio_bytes:
                    # st.write("**"+tr("Vendeur")+" :**\n")
                    # Fonction pour générer et jouer le texte en speech
                    st.audio(audio_bytes, format="audio/wav", autoplay=False)
                    try:
                        detection = False
                        if detection:
                            # Create a BytesIO object from the audio stream
                            audio_stream_bytesio = io.BytesIO(audio_bytes)

                            # Read the WAV stream using wavio
                            wav = wavio.read(audio_stream_bytesio) 

                            # Extract the audio data from the wavio.Wav object
                            audio_data = wav.data

                            # Convert the audio data to a NumPy array
                            audio_input = np.array(audio_data, dtype=np.float32)
                            audio_input = np.mean(audio_input, axis=1)/32768
                                    
                            result = model_speech.transcribe(audio_input)
                            Lang_detected = result["language"]
                            query = result["text"]
                        
                        else:
                            # Avec l'aide de la bibliothèque speech_recognition de Google
                            Lang_detected = st.session_state['Language'] 
                            # Transcription google
                            audio_stream = sr.AudioData(audio_bytes, 32000, 2) 
                            r = sr.Recognizer()
                            query = r.recognize_google(audio_stream, language = Lang_detected)
                                
                        # Transcription 
                        # st.write("**"+tr("Vendeur :")+"** "+query)
                        with st.chat_message("user"):
                            st.markdown(query)
                        st.write("")
                    
                        if query != "":
                            input_messages = [HumanMessage(query)]
                            output = app.invoke(
                                {"messages": input_messages, "language": language},
                                config,
                            )
                            #with st.chat_message("user"):
                            # Add user message to chat history
                            st.session_state.messages.append({"role": "user", "content": query})

                            # Récupération de la réponse
                            custom_sentence = output["messages"][-1].content
                            
                            # Joue l'audio
                            play_audio(custom_sentence,Lang_detected , 1)
                            
                            # st.write("**"+tr("Prospect :")+"** "+custom_sentence)
                            with st.chat_message("assistant"):
                                st.markdown(custom_sentence)
                            
                            # Add user message to chat history
                            st.session_state.messages.append({"role": "assistant", "content": custom_sentence})
                            

                            
                    except KeyboardInterrupt:
                        st.write(tr("Arrêt de la reconnaissance vocale."))
                    except:
                        st.write(tr("Problème, essayer de nouveau.."))
                st.write("")
        # Ajouter un espace pour séparer les zones
        # st.divider()
            with col2:
                if ("messages" in st.session_state) :
                    if (st.session_state.messages != []):
                    # Display chat messages from history on app rerun
                        for message in st.session_state.messages:
                            with st.chat_message(message["role"]):
                                st.markdown(message["content"])
    else:
        if to_init and not initialized:
            init_run()
        st.write("**thread_id:** "+thread_id)
        for i in range(8,len(question)):
            st.write("")

            q = st.text_input(label=".", value=tr(question[i]),label_visibility="collapsed")
            if (q !=""):
                input_messages = [HumanMessage(q)]
                output = app.invoke(
                    {"messages": input_messages, "language": language},
                    config,
                    )
            # output = app.invoke(
            #     {"messages": q,"language": language},
            #     config,
            # ) 
                custom_sentence = output["messages"][-1].content  
                st.write(custom_sentence)
                st.write("")
                if (used_model[:3] == 'mis'):
                    time.sleep(2)
                
                st.divider()