app.py

import torch
from transformers import AutoProcessor, DiaForConditionalGeneration
import gradio as gr

# Device: use GPU if available
device = "cuda" if torch.cuda.is_available() else "cpu"

# Load model and processor
checkpoint = "nari-labs/Dia-1.6B-0626"
processor = AutoProcessor.from_pretrained(checkpoint)
model = DiaForConditionalGeneration.from_pretrained(checkpoint).to(device)

def tts_dialogue(script: str):
    """
    Expects script formatted with [S1], [S2] tags for dialogue.
    Example: "[S1] Hello there! [S2] Hi, how are you?"
    """
    inputs = processor(text=script, return_tensors="pt", padding=True).to(device)
    outputs = model.generate(
        **inputs,
        max_new_tokens=3072,
        guidance_scale=3.0,
        temperature=1.8,
        top_p=0.9,
        top_k=45
    )
    audio_list = processor.batch_decode(outputs)  # returns list of audio bytes
    return (audio_list[0],)

iface = gr.Interface(
    fn=tts_dialogue,
    inputs=gr.Textbox(label="Dialogue Script", placeholder="[S1] Hello [S2] Hi!"),
    outputs=gr.Audio(label="Generated Audio"),
    title="📢 Dia 1.6B TTS Dialogue Demo",
    description="A demo using Dia 1.6B for expressive, multi‑speaker TTS"
)

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