app.py

import tempfile

from pathlib import Path

import gradio as gr
import numpy as np
import soundfile as sf
import torch

from dia.model import Dia

MODEL_ID = "nari-labs/Dia-1.6B-0626"
SAMPLE_RATE = 44100

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

model = Dia.from_pretrained(MODEL_ID, device=device)


def generate_audio(text, audio_prompt):
    if not text or text.isspace():
        raise gr.Error("Text input cannot be empty.")

    prompt_path = None

    if audio_prompt is not None:
        sr, audio = audio_prompt

        if audio is not None and audio.size > 0:
            with tempfile.NamedTemporaryFile(mode="wb", suffix=".wav", delete=False) as f:
                sf.write(f.name, audio, sr)
                prompt_path = f.name

    audio_out = model.generate(
        text=text,
        audio_prompt=prompt_path,
        max_tokens=None,
        cfg_scale=3.0,
        temperature=1.8,
        top_p=0.95,
    )

    if prompt_path is not None and Path(prompt_path).exists():
        try:
            Path(prompt_path).unlink()
        except OSError:
            pass

    audio_np = np.asarray(audio_out, dtype=np.float32)

    return SAMPLE_RATE, audio_np


with gr.Blocks() as demo:
    gr.Markdown("# Dia 1.6B-0626 Text-to-Speech")

    with gr.Row():
        with gr.Column(scale=1):
            text_in = gr.Textbox(
                label="Input text",
                lines=6,
                placeholder="Start with [S1] / [S2] tags, e.g.:\n[S1] Hello. [S2] Hi there.",
            )

            audio_prompt_in = gr.Audio(
                label="Audio prompt (optional, voice cloning)",
                sources=["upload", "microphone"],
                type="numpy",
            )

            btn = gr.Button("Generate", variant="primary")

        with gr.Column(scale=1):
            audio_out = gr.Audio(
                label="Generated audio",
                type="numpy",
                autoplay=False,
            )

    btn.click(
        fn=generate_audio,
        inputs=[text_in, audio_prompt_in],
        outputs=[audio_out],
        api_name="generate",
    )

if __name__ == "__main__":
    demo.launch()