app.py

import asyncio
import edge_tts
import gradio as gr
import os
from datetime import datetime
from pydub import AudioSegment

AUDIO_DIR = "saved_audios"
PODCAST_DIR = "podcast_audios"
os.makedirs(AUDIO_DIR, exist_ok=True)
os.makedirs(PODCAST_DIR, exist_ok=True)

def generate_unique_filename(folder, prefix="audio", ext="mp3"):
    timestamp = datetime.now().strftime("%Y%m%d_%H%M%S_%f")
    return os.path.join(folder, f"{prefix}_{timestamp}.{ext}")

async def generate_speech(text, voice, rate, pitch, folder=AUDIO_DIR):
    output_file = generate_unique_filename(folder)
    communicate = edge_tts.Communicate(text, voice, rate=rate, pitch=pitch)
    await communicate.save(output_file)
    return output_file

async def get_voices():
    voices = await edge_tts.list_voices()
    return [voice["ShortName"] for voice in voices]

def list_saved_audios():
    files = sorted(os.listdir(AUDIO_DIR), reverse=True)
    return [os.path.join(AUDIO_DIR, f) for f in files if f.endswith(".mp3")]

def list_saved_podcasts():
    files = sorted(os.listdir(PODCAST_DIR), reverse=True)
    return [os.path.join(PODCAST_DIR, f) for f in files if f.endswith(".mp3")]

async def tts_interface(text, voice, rate_percentage, pitch_hz):
    rate = f"{'+' if rate_percentage >= 0 else ''}{rate_percentage}%"
    pitch = f"{'+' if pitch_hz >= 0 else ''}{pitch_hz}Hz"
    audio_path = await generate_speech(text, voice, rate, pitch)
    return audio_path

def play_saved_audio(audio_file):
    return audio_file

async def podcast_produce(scripts, voice, rate_percentage, pitch_hz, bgm_file, podcast_title, podcast_desc):
    rate = f"{'+' if rate_percentage >= 0 else ''}{rate_percentage}%"
    pitch = f"{'+' if pitch_hz >= 0 else ''}{pitch_hz}Hz"
    audio_segments = []
    for idx, text in enumerate(scripts):
        if text.strip():
            temp_audio = generate_unique_filename(PODCAST_DIR, prefix=f"segment{idx}")
            communicate = edge_tts.Communicate(text, voice, rate=rate, pitch=pitch)
            await communicate.save(temp_audio)
            audio_segments.append(AudioSegment.from_file(temp_audio))
            os.remove(temp_audio)
    if not audio_segments:
        return None
    podcast_audio = sum(audio_segments)
    if bgm_file is not None and hasattr(bgm_file, "name") and os.path.isfile(bgm_file.name):
        bgm = AudioSegment.from_file(bgm_file.name).apply_gain(-10)
        bgm = bgm[:len(podcast_audio)]
        podcast_audio = podcast_audio.overlay(bgm)
    podcast_file = generate_unique_filename(PODCAST_DIR, prefix="podcast")
    podcast_audio.export(podcast_file, format="mp3")
    meta_file = podcast_file.replace(".mp3", ".txt")
    with open(meta_file, "w", encoding="utf-8") as f:
        f.write(f"Title: {podcast_title}\nDescription: {podcast_desc}\n")
    return podcast_file

def clear_textbox():
    return ""

def clear_paragraphs():
    return [""]

def add_paragraph(paragraphs):
    paragraphs = paragraphs.copy()
    paragraphs.append("")
    return paragraphs

def remove_paragraph(paragraphs):
    paragraphs = paragraphs.copy()
    if len(paragraphs) > 1:
        paragraphs.pop()
    return paragraphs

def update_paragraphs_ui(paragraphs):
    # 回傳一組 Textbox 元件
    return [gr.Textbox(value=p, label=f"段落{i+1}內容", lines=3, interactive=True) for i, p in enumerate(paragraphs)]

def collect_paragraphs(*args):
    # 收集所有段落內容
    return list(args)

async def main():
    voices = await get_voices()
    with gr.Blocks(theme=gr.themes.Soft()) as demo:
        gr.Markdown("## 🎙️ Edge TTS 語音合成與播客製作\n\n- 多段腳本自由增減、內容可清空\n- 介面直覺、操作友善、檔案自動管理")

        with gr.Tab("語音合成"):
            with gr.Row():
                text_input = gr.Textbox(lines=5, label="輸入文本")
                clear_btn = gr.Button("清空")
            voice_input = gr.Dropdown(voices, label="選擇語音", value="zh-CN-XiaoxiaoNeural")
            rate_input = gr.Slider(-50, 50, value=0, step=1, label="語速調整 (%)")
            pitch_input = gr.Slider(-50, 50, value=0, step=1, label="音高調整 (Hz)")
            tts_btn = gr.Button("生成語音")
            audio_output = gr.Audio(type="filepath", label="生成的語音")
            tts_btn.click(
                fn=tts_interface,
                inputs=[text_input, voice_input, rate_input, pitch_input],
                outputs=audio_output
            )
            clear_btn.click(fn=clear_textbox, outputs=text_input)

        with gr.Tab("檢視已儲存語音"):
            audio_files = gr.Dropdown(list_saved_audios(), label="選擇已儲存語音檔案", interactive=True)
            saved_audio_output = gr.Audio(type="filepath", label="播放已儲存語音")
            audio_files.change(fn=play_saved_audio, inputs=audio_files, outputs=saved_audio_output)

        with gr.Tab("播客製作"):
            gr.Markdown("### 📝 多段腳本輸入（可自由增減段落）")
            paragraphs_state = gr.State([""])
            paragraphs_container = gr.Column()

            # 初始渲染
            paragraph_boxes = update_paragraphs_ui([""])
            for tb in paragraph_boxes:
                paragraphs_container.append(tb)

            add_btn = gr.Button("新增段落")
            remove_btn = gr.Button("刪除段落")
            clear_all_btn = gr.Button("全部清空")

            def on_add(paragraphs):
                new_paragraphs = add_paragraph(paragraphs)
                return new_paragraphs, gr.update(components=update_paragraphs_ui(new_paragraphs))

            def on_remove(paragraphs):
                new_paragraphs = remove_paragraph(paragraphs)
                return new_paragraphs, gr.update(components=update_paragraphs_ui(new_paragraphs))

            def on_clear():
                new_paragraphs = clear_paragraphs()
                return new_paragraphs, gr.update(components=update_paragraphs_ui(new_paragraphs))

            add_btn.click(on_add, inputs=paragraphs_state, outputs=[paragraphs_state, paragraphs_container])
            remove_btn.click(on_remove, inputs=paragraphs_state, outputs=[paragraphs_state, paragraphs_container])
            clear_all_btn.click(on_clear, outputs=[paragraphs_state, paragraphs_container])

            # 參數設定
            voice_input2 = gr.Dropdown(voices, label="選擇語音", value="zh-CN-XiaoxiaoNeural")
            rate_input2 = gr.Slider(-50, 50, value=0, step=1, label="語速調整 (%)")
            pitch_input2 = gr.Slider(-50, 50, value=0, step=1, label="音高調整 (Hz)")
            bgm_input = gr.File(label="上傳背景音樂（可選）")
            podcast_title = gr.Textbox(label="播客標題")
            podcast_desc = gr.Textbox(label="播客描述")
            podcast_btn = gr.Button("生成播客")
            podcast_output = gr.Audio(type="filepath", label="生成的播客音檔")

            def gather_scripts(*args):
                return list(args)

            def on_podcast_btn_click(*args):
                # args: 段落內容 + 參數
                n = len(paragraphs_state.value)
                scripts = list(args[:n])
                voice = args[n]
                rate = args[n+1]
                pitch = args[n+2]
                bgm = args[n+3]
                title = args[n+4]
                desc = args[n+5]
                return asyncio.run(podcast_produce(scripts, voice, rate, pitch, bgm, title, desc))

            # 這裡需要用最新的段落數來組合 inputs
            def get_inputs():
                return [tb for tb in paragraphs_container.children] + [voice_input2, rate_input2, pitch_input2, bgm_input, podcast_title, podcast_desc]

            podcast_btn.click(
                fn=on_podcast_btn_click,
                inputs=lambda: get_inputs(),
                outputs=podcast_output
            )

        with gr.Tab("檢視已儲存播客"):
            podcast_files = gr.Dropdown(list_saved_podcasts(), label="選擇已儲存播客檔案", interactive=True)
            saved_podcast_output = gr.Audio(type="filepath", label="播放已儲存播客")
            podcast_files.change(fn=play_saved_audio, inputs=podcast_files, outputs=saved_podcast_output)

    demo.launch()

if __name__ == "__main__":
    asyncio.run(main())