app.py

import gradio as gr
from kokoro import KPipeline
import soundfile as sf
import numpy as np
import os
import datetime
from tqdm import tqdm

# --- Configuration ---
SAVE_DIR = "saved_audios"
os.makedirs(SAVE_DIR, exist_ok=True)

# --- 官方語音列表（可根據官方文檔擴充）---
VOICE_PRESETS = {
    "美式英語 (女聲)": {'lang_code': 'a', 'voice': 'af_heart'},
    "美式英語 (男聲)": {'lang_code': 'a', 'voice': 'am_adam'},
    "英式英語 (女聲)": {'lang_code': 'b', 'voice': 'bf_emma'},
    "英式英語 (男聲)": {'lang_code': 'b', 'voice': 'bm_george'},
    "簡體中文 (女聲 - 小貝)": {'lang_code': 'z', 'voice': 'zf_xiaobei'},
    "簡體中文 (女聲 - 小妮)": {'lang_code': 'z', 'voice': 'zf_xiaoni'},
    "繁體中文 (女聲 - 小貝)": {'lang_code': 'z', 'voice': 'zf_xiaobei'},
}

# --- 全局變數 ---
global_pipeline = None
current_lang_code = None

def get_saved_files_dropdown_options():
    files = [f for f in os.listdir(SAVE_DIR) if f.endswith(".wav")]
    files.sort(key=lambda f: os.path.getmtime(os.path.join(SAVE_DIR, f)), reverse=True)
    return files

def synthesize_speech_gradio(text, preset_name):
    global global_pipeline, current_lang_code
    status_message = ""
    if not text:
        status_message = "請輸入文字！"
        gr.Warning(status_message)
        return status_message, None, gr.update(choices=get_saved_files_dropdown_options()), gr.update(value=None)
    preset = VOICE_PRESETS[preset_name]
    lang_code = preset['lang_code']
    voice = preset['voice']
    if global_pipeline is None or current_lang_code != lang_code:
        status_message = f"正在初始化 Kokoro TTS 管線 (語言: {lang_code})..."
        try:
            global_pipeline = KPipeline(lang_code=lang_code, repo_id='hexgrad/Kokoro-82M')
            current_lang_code = lang_code
            status_message = "模型初始化成功。"
        except Exception as e:
            status_message = f"模型初始化失敗: {e}"
            gr.Error(status_message)
            return status_message, None, gr.update(choices=get_saved_files_dropdown_options()), gr.update(value=None)
    status_message = f"正在生成語音 (預設: {preset_name})..."
    audio_segments = []
    sample_rate = 24000
    try:
        generator = global_pipeline(text, voice=voice)
        segment_count = 0
        print("語音片段生成進度：")
        for i, (graphemes, phonemes, audio) in enumerate(tqdm(generator, desc="生成中", unit="段")):
            audio_segments.append(audio)
            segment_count += 1
            # 即時顯示每段的 graphemes 和 phonemes
            print(f"片段 {i+1}:")
            print(f"  Graphemes: {graphemes}")
            print(f"  Phonemes:  {phonemes}")
        print()  # 進度條結束後換行
        status_message = f"所有 {segment_count} 個語音片段生成完成。"
    except Exception as e:
        status_message = f"語音生成失敗，請檢查 voice 名稱是否正確: {e}"
        gr.Error(status_message)
        return status_message, None, gr.update(choices=get_saved_files_dropdown_options()), gr.update(value=None)
    if not audio_segments:
        status_message = "沒有生成任何音訊片段。"
        gr.Warning(status_message)
        return status_message, None, gr.update(choices=get_saved_files_dropdown_options()), gr.update(value=None)
    full_audio = np.concatenate(audio_segments)
    timestamp = datetime.datetime.now().strftime("%Y%m%d_%H%M%S")
    safe_text_prefix = "".join(c for c in text[:20] if c.isalnum() or c.isspace()).strip()
    if safe_text_prefix:
        filename = f"kokoro_{safe_text_prefix}_{timestamp}.wav"
    else:
        filename = f"kokoro_audio_{timestamp}.wav"
    file_path = os.path.join(SAVE_DIR, filename)
    try:
        sf.write(file_path, full_audio, sample_rate)
        status_message = f"語音生成成功並已保存為 {filename}。"
        gr.Info(status_message)
    except Exception as e:
        status_message = f"語音生成成功但保存失敗: {e}"
        gr.Error(status_message)
    return status_message, (sample_rate, full_audio), gr.update(choices=get_saved_files_dropdown_options()), gr.update(value=None)

def select_file_for_download(filename):
    if filename:
        file_path = os.path.join(SAVE_DIR, filename)
        if os.path.exists(file_path):
            return file_path
    return None

def delete_selected_audio(filename_to_delete):
    status_message = ""
    if filename_to_delete:
        file_path = os.path.join(SAVE_DIR, filename_to_delete)
        if os.path.exists(file_path):
            os.remove(file_path)
            status_message = f"已刪除 {filename_to_delete}"
            gr.Info(status_message)
            return status_message, gr.update(choices=get_saved_files_dropdown_options()), gr.update(value=None)
        else:
            status_message = f"檔案 {filename_to_delete} 不存在。"
            gr.Warning(status_message)
            return status_message, gr.update(choices=get_saved_files_dropdown_options()), gr.update(value=None)
    status_message = "請選擇要刪除的檔案。"
    gr.Warning(status_message)
    return status_message, gr.update(choices=get_saved_files_dropdown_options()), gr.update(value=None)

def clear_interface():
    return "", None, "", gr.update(choices=get_saved_files_dropdown_options()), gr.update(value=None)

with gr.Blocks() as demo:
    gr.Markdown("""
    # Kokoro TTS 語音合成器
    在 Hugging Face Spaces 上使用 Kokoro TTS 生成高品質語音。
    **注意：繁體中文選項使用簡體中文模型，發音可能不完全符合台灣國語腔調。**
    **重要提示：此應用程式在 Hugging Face Spaces 上運行，生成的音檔會保存到伺服器，但 Space 重啟後可能會丟失。**
    """)
    with gr.Tab("語音生成"):
        with gr.Row():
            with gr.Column():
                text_input = gr.Textbox(lines=5, label="輸入文字", placeholder="請輸入您想轉換為語音的文字...")
                voice_dropdown = gr.Dropdown(list(VOICE_PRESETS.keys()), label="選擇語音預設", value="美式英語 (女聲)")
                with gr.Row():
                    generate_btn = gr.Button("生成語音")
                    clear_btn = gr.Button("清除")
            with gr.Column():
                audio_output = gr.Audio(label="生成的語音", type="numpy")
                gr.Markdown("**提示：** 音訊播放器右下角有下載按鈕，點擊即可保存音檔到您的電腦。")
                info_output = gr.Textbox(label="狀態/提示", interactive=False)
    with gr.Tab("管理已保存音檔"):
        gr.Markdown("### 已保存的音檔列表")
        saved_files_dropdown = gr.Dropdown(
            choices=get_saved_files_dropdown_options(),
            label="選擇音檔",
            info="選擇一個音檔進行下載或刪除"
        )
        download_file_output = gr.File(label="下載選定音檔", interactive=False)
        with gr.Row():
            refresh_files_btn = gr.Button("刷新列表")
            delete_btn = gr.Button("刪除選定音檔")
        delete_info_output = gr.Textbox(label="刪除狀態", interactive=False)
    generate_btn.click(
        fn=synthesize_speech_gradio,
        inputs=[text_input, voice_dropdown],
        outputs=[info_output, audio_output, saved_files_dropdown, download_file_output]
    )
    clear_btn.click(
        fn=clear_interface,
        inputs=[],
        outputs=[text_input, audio_output, info_output, saved_files_dropdown, download_file_output]
    )
    saved_files_dropdown.change(
        fn=select_file_for_download,
        inputs=[saved_files_dropdown],
        outputs=[download_file_output]
    )
    refresh_files_btn.click(
        fn=lambda: ("", gr.update(choices=get_saved_files_dropdown_options()), gr.update(value=None)),
        inputs=[],
        outputs=[delete_info_output, saved_files_dropdown, download_file_output]
    )
    delete_btn.click(
        fn=delete_selected_audio,
        inputs=[saved_files_dropdown],
        outputs=[delete_info_output, saved_files_dropdown, download_file_output]
    )
demo.launch()