app.py

import gradio as gr
from voxcpm import VoxCPM
import soundfile as sf
import tempfile
import numpy as np

model = None

def load_model():
    global model
    if model is None:
        model = VoxCPM.from_pretrained("openbmb/VoxCPM2", load_denoiser=False)
    return model

def generate_speech(text, ref_audio, mode):
    m = load_model()

    if mode == "🎨 Voice Design" or ref_audio is None:
        # TTS biasa / voice design dari deskripsi
        wav = m.generate(text=text, cfg_value=2.0, inference_timesteps=10)
    else:
        # Voice cloning pakai audio referensi
        ref_sr, ref_wav = ref_audio
        ref_wav = ref_wav.astype(np.float32) / 32768.0
        wav = m.generate(
            text=text,
            ref_audio=ref_wav,
            ref_sr=ref_sr,
            cfg_value=2.0,
            inference_timesteps=10,
        )

    tmp = tempfile.NamedTemporaryFile(suffix=".wav", delete=False)
    sf.write(tmp.name, wav, m.tts_model.sample_rate)
    return tmp.name

demo = gr.Interface(
    fn=generate_speech,
    inputs=[
        gr.Textbox(
            label="Teks yang mau diucapkan",
            placeholder="Ketik teks di sini... (untuk Voice Design, awali dengan (deskripsi suara)teks)"
        ),
        gr.Audio(
            label="🎙️ Upload audio referensi (untuk Voice Cloning)",
            type="numpy",
            sources=["upload", "microphone"]
        ),
        gr.Radio(
            choices=["🎨 Voice Design", "🎙️ Voice Cloning"],
            value="🎙️ Voice Cloning",
            label="Mode"
        ),
    ],
    outputs=gr.Audio(label="Hasil suara"),
    title="VoxCPM TTS API",
    description="**Voice Cloning:** upload audio referensi + ketik teks\n**Voice Design:** tulis (deskripsi suara)teks, tanpa audio referensi"
)

demo.launch()