app.py

import gradio as gr
import base64
import requests
import secrets
import os
import argparse

from io import BytesIO
from pydub import AudioSegment


LOCAL_API_ENDPOINT = "http://localhost:5000"
PUBLIC_API_ENDPOINT = "http://121.176.153.117:5000"

API_ENDPOINT = PUBLIC_API_ENDPOINT

session_id = ""
chat_history = []

css = """
#audio_input {
    margin-top: -30px; !important;
    margin-left: -15px; !important;
    width: 100% !important;
}

#audio_input button {
    height:50px !important;
    font-size: 0px !important;
    width: 110% !important;
}

#audio_input button:after {
    content: '🎤' !important;
    font-size: 16px !important;
}

audio {
    min-width: 200px !important;
}

@media (max-width : 480px) {
    #audio_input {
        width: 120% !important;
    }

    #audio_input button:after {
        content: '' !important;
    }

    #txt_input_container {
        flex-grow: 70% !important;
    }

    #audio_input_container {
        flex-grow: 30% !important;
    }
}


"""

js_audio_auto_play = """
() => {
    // select last audio element
    const audio = document.getElementsByTagName('audio');
    const last_audio = audio[audio.length - 1];

    // set autoplay attribute
    last_audio.setAttribute('autoplay', true);
}
"""

def create_chat_session():
    r = requests.post(API_ENDPOINT + "/create")
        
    if (r.status_code != 201):
        raise Exception("Failed to create chat session")
    
    # create temp audio folder
    session_id = r.json()["id"]
    os.makedirs(f"./temp_audio/{session_id}")
        
    return session_id

def create_new_or_change_session(history, id):
    global session_id
    global chat_history

    if id == "":
        session_id = create_chat_session()
        history = []
    else:
        history, _ = change_session(history, id)

    chat_history = history

    return history, gr.update(value="", interactive=False)

def add_text(history, text):
    history = history + [(text, None)]
    return history, gr.update(value="", interactive=False)


def add_audio(history, audio):

    audio_bytes = base64.b64decode(audio['data'].split(',')[-1].encode('utf-8'))

    audio_file = BytesIO(audio_bytes)
    AudioSegment.from_file(audio_file).export(audio_file, format="mp3")

    # save audio file temporary to disk
    audio_id = secrets.token_hex(8)
    AudioSegment.from_file(audio_file).export(f"temp_audio/{session_id}/audio_input_{audio_id}.mp3", format="mp3")

    history = history + [((f"temp_audio/{session_id}/audio_input_{audio_id}.mp3",), None)]

    response = requests.post(
        API_ENDPOINT + "/transcribe",
        files={'audio': audio_file.getvalue()}
    )

    if (response.status_code != 200):
        raise Exception(response.text)
    
    text = response.json()['text']

    history = history + [(text, None)]

    return history, gr.update(value="", interactive=False)

def reset_chat_session(history):
    global session_id
    global chat_history

    response = requests.post(
        API_ENDPOINT + f"/reset/{session_id}"
    )

    if (response.status_code != 200):
        raise Exception(response.text)
    
    history = []
    chat_history = []

    return history

def bot(history):
    if type(history[-1][0]) == str:
        message = history[-1][0]
    else:
        message = history[-2][0]

    response = requests.post(
        API_ENDPOINT + f"/send/text/{session_id}",
        headers={'Content-type': 'application/json'},
        json={
            'message': message,
            'role': 'user'
        }
    )

    if (response.status_code != 200):
        raise Exception(f"Failed to send message, {response.text}")
    
    response = response.json()

    text, audio = response['text'], response['audio']

    audio_bytes = base64.b64decode(audio.encode('utf-8'))
    audio_file = BytesIO(audio_bytes)
    audio_id = secrets.token_hex(8)
    AudioSegment.from_file(audio_file).export(f"temp_audio/{session_id}/audio_input_{audio_id}.mp3", format="mp3")

    history = history + [(None, (f"temp_audio/{session_id}/audio_input_{audio_id}.mp3",))]
    history = history + [(None, text)]
    
    global chat_history

    chat_history = history.copy()

    return history

def change_session(history, id):
    global session_id
    global chat_history

    response = requests.get(
        API_ENDPOINT + f"/{id}"
    )

    if (response.status_code != 200):
        raise Exception(response.text)
    
    response = response.json()

    session_id = id

    history = []

    try:
        for chat in response:
            if chat['role'] == 'user':
                if chat['audio'] != "":
                    audio_bytes = base64.b64decode(chat['audio'].encode('utf-8'))
                    audio_file = BytesIO(audio_bytes)
                    audio_id = secrets.token_hex(8)
                    AudioSegment.from_file(audio_file).export(f"temp_audio/{id}/audio_input_{audio_id}.mp3", format="mp3")
                    history = history + [((f"temp_audio/{id}/audio_input_{audio_id}.mp3",), None)]
                history = history + [(chat['message'], None)]
            elif chat['role'] == 'assistant':
                audio_bytes = base64.b64decode(chat['audio'].encode('utf-8'))
                audio_file = BytesIO(audio_bytes)
                audio_id = secrets.token_hex(8)
                AudioSegment.from_file(audio_file).export(f"temp_audio/{id}/audio_input_{audio_id}.mp3", format="mp3")

                history = history + [(None, (f"temp_audio/{id}/audio_input_{audio_id}.mp3",))]
                history = history + [(None, chat['message'])]
            else:
                raise Exception("Invalid chat role")
    except Exception as e:
        raise Exception(f"Response: {response}")

    chat_history = history.copy()

    print(f"len(chat_history): {len(chat_history)}\nlen(history): {len(history)}\nlen(response): {len(response)}")

    return history, gr.update(value="", interactive=False)

def load_chat_history(history):
    global chat_history
    if len(chat_history) > len(history):
        history = chat_history
    return history

def main():
    global session_id
    global chat_history

    session_id = create_chat_session()
    chat_history = []

    with gr.Blocks(css=css) as demo:
        with gr.Row():
            # change session id
            change_session_txt = gr.Textbox(
                show_label=False,
                placeholder=session_id,
            ).style(container=False)
        with gr.Row():
            # button to create new or change session id
            change_session_button = gr.Button(
                "Create new or change session", type='success', size="sm"
            ).style(container=False)

        chatbot = gr.Chatbot([], elem_id="chatbot").style(height=750)

        demo.load(load_chat_history, [chatbot], [chatbot], queue=False)

        with gr.Row():
            with gr.Column(scale=0.85, min_width=0, elem_id="txt_input_container"):
                txt = gr.Textbox(
                    show_label=False,
                    placeholder="Enter text and press enter, or record audio",
                    elem_id="txt_input"
                ).style(container=False)
            with gr.Column(scale=0.15, min_width=0, elem_id="audio_input_container"):
                audio = gr.Audio(
                    source="microphone", type="numpy", show_label=False, format="mp3", min_width=0, container=False, elem_id="audio_input"
                )

        with gr.Row():
            reset_button = gr.Button(
                "Reset Chat Session", type='stop', size="sm"
            ).style(container=False)

        txt_msg = txt.submit(add_text, [chatbot, txt], [chatbot, txt], queue=False).then(
            bot, chatbot, chatbot
        ).then(
            None, [], [], queue=False, _js=js_audio_auto_play
        )
        txt_msg.then(lambda: gr.update(interactive=True), None, [txt], queue=False)

        audio_msg = audio.change(add_audio, [chatbot, audio], [chatbot, audio], queue=False, preprocess=False, postprocess=False).then(
            bot, chatbot, chatbot
        ).then(
            None, [], [], queue=False, _js=js_audio_auto_play
        )
        audio_msg.then(lambda: gr.update(interactive=True, value=None), None, [audio], queue=False)

        reset_button.click(reset_chat_session, [chatbot], [chatbot], queue=False)

        chgn_msg = change_session_txt.submit(change_session, [chatbot, change_session_txt], [chatbot, change_session_txt], queue=False)
        chgn_msg.then(lambda: gr.update(interactive=True, placeholder=session_id), None, [change_session_txt], queue=False)

        create_new_or_change_session_btn = change_session_button.click(create_new_or_change_session, [chatbot, change_session_txt], [chatbot, change_session_txt], queue=False)
        create_new_or_change_session_btn.then(lambda: gr.update(interactive=True, placeholder=session_id), None, [change_session_txt], queue=False)

        return demo


if __name__ == "__main__":
    # arguments --local
    parser = argparse.ArgumentParser()
    parser.add_argument("--local", action="store_true", help="Use local API endpoint")
    args = parser.parse_args()

    if args.local:
        API_ENDPOINT = LOCAL_API_ENDPOINT

    demo = main()
    demo.launch(show_error=True, server_name="0.0.0.0")