app.py

"""
🎙️ Kokoro-TTS-only demo – Zero-GPU edition
Routes every synthesis to an idle A100.
"""

import os, tempfile, subprocess, numpy as np
import gradio as gr
import spaces           # Zero-GPU decorator
import soundfile as sf

# ------------------------------------------------------------------
# 1.  Lazy Kokoro loader (runs once per GPU worker)
# ------------------------------------------------------------------
kokoro_pipe = None

def load_kokoro():
    global kokoro_pipe
    if kokoro_pipe is None:
        from kokoro import KPipeline
        kokoro_pipe = KPipeline(lang_code='a')
    return kokoro_pipe

# ------------------------------------------------------------------
# 2.  Generation helper
# ------------------------------------------------------------------
@spaces.GPU
def tts_kokoro(text, voice, speed):
    pipe = load_kokoro()
    generator = pipe(text, voice=voice, speed=speed)
    for gs, ps, audio in generator:
        return audio
    raise RuntimeError("Kokoro generation failed")

# ------------------------------------------------------------------
# 3.  Zero-GPU entry point
# ------------------------------------------------------------------
@spaces.GPU
def synthesise(text, voice, speed):
    if not text.strip():
        raise gr.Error("Please enter some text.")
    wav = tts_kokoro(text, voice=voice, speed=speed)
    fd, tmp = tempfile.mkstemp(suffix='.wav')
    os.close(fd)
    sf.write(tmp, wav, 24000)
    return tmp

# ------------------------------------------------------------------
# 4.  Gradio UI
# ------------------------------------------------------------------
css = """footer {visibility: hidden}"""

with gr.Blocks(css=css, title="Kokoro TTS – Zero-GPU") as demo:
    gr.Markdown("## 🎙️ Kokoro TTS – Zero-GPU Demo")

    with gr.Row():
        with gr.Column():
            voice = gr.Dropdown(
                label="Voice",
                choices=['af_heart', 'af_sky', 'af_mist', 'af_dusk'],
                value='af_heart'
            )
            speed = gr.Slider(0.5, 2.0, 1.0, step=0.1, label="Speed")

        with gr.Column(scale=3):
            text = gr.Textbox(
                label="Text to speak",
                placeholder="Type or paste text here …",
                lines=6, max_lines=12
            )
            btn = gr.Button("🎧 Synthesise", variant="primary")
            audio_out = gr.Audio(label="Generated speech", type="filepath")

    btn.click(synthesise, inputs=[text, voice, speed], outputs=audio_out)

    gr.Markdown("### Tips  \n"
                "- **Kokoro** – fast, high-quality English TTS  \n"
                "Audio is returned as 24 kHz WAV.")

demo.launch()