app.py

import base64
import os
import shutil
import threading
import time

import gradio as gr

from api import audio_text
from blog_class import generate_item, knowledge_class, save_feedback
from design import (
    custom_css,
    custom_head,
    js_func,
    params_text,
    speaker_text,
    title_html,
    title_text,
)
from prompts import TRANS_Prompt, TRANS_Prompt_template

knowledge_cls = knowledge_class()


language_dict = {"en": "English", "cn": "Chinese"}


def update_knowledge():
    knowledge_cls.get_new_knowledge()
    choices_list = list(knowledge_cls.reference_dict.keys())

    return gr.update(choices=choices_list)


def user_message_fn(user_message, history):
    if history is None:
        history = []
    history.append({"role": "user", "content": user_message})
    return "", history


def answer_question(
    history, creative_flag, full_flag, audio_flag, speaker_flag, language_flag
):
    print("chat Start")
    answer_text, info_title, audio_file_path = knowledge_cls.get_chat_answer(
        history, creative_flag, full_flag, speaker_flag
    )
    if not answer_text:
        yield history
        return
    if language_flag not in ["jp"]:
        info_title = info_title + f"日本語返信:\n {answer_text}\n"
        sys_prompt = TRANS_Prompt.format(lang=language_dict[language_flag])
        user_prompt = TRANS_Prompt_template.format(
            japanese_text=answer_text, lang=language_dict[language_flag]
        )
        answer_text = generate_item(user_prompt, sys_prompt)

    # text
    history.append({"role": "assistant", "content": ""})
    # audio
    history.append(
        {
            "role": "assistant",
            "content": "音声なし",
            "metadata": {"title": "🎧 音声返信", "status": "done"},
        }
    )
    # ref
    history.append(
        {
            "role": "assistant",
            "content": info_title,
            "metadata": {"title": "📚 参照ナレッジ", "status": "done"},
        }
    )

    def run_audio_answer():
        run_audio_path = audio_text(answer_text, audio_file_path, speaker_flag)
        audio_str = base64.b64encode(open(run_audio_path, "rb").read()).decode("utf-8")
        history[-2]["content"] = gr.HTML(
            f"""
            <audio controls>
                <source src="data:audio/mp3;base64,{audio_str}" type="audio/mp3">
            </audio>
            """
        )
        history[-2]["metadata"] = {"title": "🎧 音声返信", "status": "done"}

    if not audio_flag:
        audio_thread = threading.Thread(target=run_audio_answer)
        audio_thread.start()
        history[-2]["content"] = "音声作成中..."
        history[-2]["metadata"]["status"] = "pending"

    #

    for char in answer_text:
        history[-3]["content"] += char
        yield history
        time.sleep(0.03)
    if not audio_flag:
        while audio_thread.is_alive():
            yield history
            time.sleep(0.2)
        yield history


def handle_feedback(like_data: gr.LikeData):
    save_feedback(like_data.value[0], like_data.liked)


def get_reference_info(selected_reference):
    reference_dict = knowledge_cls.reference_dict.get(selected_reference, None)
    if reference_dict:
        yield (
            reference_dict["audio"],
            reference_dict["original_text"],
            reference_dict["summary"],
        )
    else:
        yield (None, None, None)


def zip_directory():
    zip_path = "resource.zip"
    if os.path.exists(zip_path):
        os.remove(zip_path)
    shutil.make_archive(zip_path.replace(".zip", ""), "zip", "./resource")
    return zip_path


def toggle_change():
    return None, None


with gr.Blocks(title="勉強会", css=custom_css, head=custom_head, js=js_func) as demo:
    with gr.Sidebar(label="回答スタイル調整", open=False):
        with gr.Accordion("回答スタイルの違いについて", open=False):
            gr.Markdown(params_text)

        creativity_toggle = gr.Radio(
            [("創造的な回答", True), ("事実的な回答", False)],
            value=False,
            label="回答スタイル",
        )
        length_toggle = gr.Radio(
            [("詳細な説明", True), ("簡潔な説明", False)],
            value=False,
            label="説明スタイル",
        )
        audio_toggle = gr.Radio(
            [("テキスト", True), ("音声とテキスト", False)],
            value=False,
            label="返信スタイル",
        )
        language_toggle = gr.Radio(
            [("日本語", "jp"), ("中文", "cn"), ("English", "en")],
            value="jp",
            label="言語スタイル",
        )
        with gr.Accordion("話者紹介", open=False):
            gr.Markdown(speaker_text)
        speaker_toggle = gr.Radio(
            [
                ("さらとが", "matsu"),
                ("ケンシロウ", "ken"),
                ("まな", "mana"),
                ("藤井", "ren"),
                ("内山", "cai"),
            ],
            value="matsu",
            label="話者スタイル",
        )

        # creativity_toggle = Toggle(
        #     label="創意内容",
        #     value=False,
        #     interactive=True,
        #     color="#007bff",
        # )
        # length_toggle = Toggle(
        #     label="詳細内容",
        #     value=False,
        #     interactive=True,
        #     color="#007bff",
        # )
    with gr.Column(elem_id="title-area"):
        gr.HTML(title_html)
        gr.HTML(title_text)

    with gr.Column(elem_id="chatbot-container"):

        gr.HTML(
            "<h3><i class='fa fa-question-circle'></i> 何かお困りですか？</h3>",
        )

        chatbot = gr.Chatbot(
            show_label=False,
            type="messages",
            bubble_full_width=False,
            avatar_images=(
                None,
                "https://cdn.profile-image.st-hatena.com/users/saratoga623/profile.png?1728512391",
            ),
        )

        with gr.Row(elem_id="submit-btn"):
            msg = gr.Textbox(
                placeholder="質問を入力してEnter...",
                show_label=False,
                scale=5,
                container=False,
            )
            submit_btn = gr.Button(value="送信 🚀", scale=1)

            clear_btn = gr.ClearButton([msg, chatbot], value="クリア 🧹", scale=1)

    with gr.Accordion("ナレッジ一覧", elem_id="references-container", open=False):
        with gr.Column():
            with gr.Row():
                gr.HTML("<h3><i class='fa fa-book'></i> ナレッジ一覧</h3>")
                update_btn = gr.Button(value="🔄 更新", elem_id="update-btn")
            references_dropdown = gr.Dropdown(
                label="ナレッジを選ぶ",
                choices=list(knowledge_cls.reference_dict.keys()),
                value=None,
            )
            reference_audio = gr.Audio(
                label="ナレッジ要約を聞く",
                type="filepath",
                interactive=False,
                waveform_options=gr.WaveformOptions(
                    show_recording_waveform=False,
                    show_controls=False,
                ),
            )
            sum_references_box = gr.Textbox(
                label="ナレッジ要約",
                lines=4,
                interactive=False,
            )
            with gr.Accordion("ナレッジ详情", open=False):
                all_references_box = gr.Markdown(max_height=250)

    with gr.Accordion("資源", open=False, elem_id="download-container"):
        download_btn = gr.Button("ダウンロード", elem_id="update-btn")
        file_output = gr.File(show_label=False, container=False)

    msg.submit(
        fn=user_message_fn, inputs=[msg, chatbot], outputs=[msg, chatbot], queue=False
    ).then(
        fn=answer_question,
        inputs=[
            chatbot,
            creativity_toggle,
            length_toggle,
            audio_toggle,
            speaker_toggle,
            language_toggle,
        ],
        outputs=chatbot,
    )

    submit_btn.click(
        fn=user_message_fn, inputs=[msg, chatbot], outputs=[msg, chatbot], queue=False
    ).then(
        fn=answer_question,
        inputs=[
            chatbot,
            creativity_toggle,
            length_toggle,
            audio_toggle,
            speaker_toggle,
            language_toggle,
        ],
        outputs=chatbot,
    )
    chatbot.like(handle_feedback, None, None, like_user_message=True)
    update_btn.click(
        fn=update_knowledge,
        outputs=references_dropdown,
    )

    references_dropdown.change(
        fn=get_reference_info,
        inputs=references_dropdown,
        outputs=[reference_audio, all_references_box, sum_references_box],
    )

    download_btn.click(fn=zip_directory, outputs=file_output)
    creativity_toggle.change(fn=toggle_change, inputs=None, outputs=[msg, chatbot])
    length_toggle.change(fn=toggle_change, inputs=None, outputs=[msg, chatbot])
    audio_toggle.change(fn=toggle_change, inputs=None, outputs=[msg, chatbot])
    language_toggle.change(fn=toggle_change, inputs=None, outputs=[msg, chatbot])
    speaker_toggle.change(fn=toggle_change, inputs=None, outputs=[msg, chatbot])

# app
if __name__ == "__main__":
    demo.launch(inline=False, share=False, debug=True)