app.py

import gradio as gr
from transformers import pipeline

# Maya1 TTS pipeline
pipe = pipeline("text-to-speech", model="maya-research/maya1", device="cuda")

def generate_tts(text, voice_description):
    # Combine text and the voice style prompt
    full_prompt = f"{voice_description} {text}"

    out = pipe(full_prompt)
    audio = out["audio"]

    # The pipeline outputs sample rate = 44100 by default
    return (44100, audio)

# Default Torch Em voice description
DEFAULT_VOICE = "<male><40yo><deep><baritone><stern><aggressive>"

with gr.Blocks() as demo:
    gr.Markdown("## 🔥 Maya1 TTS Demo for Torch Em (GPU Required)")

    text = gr.Textbox(
        label="Text Input",
        value="This is Torch Em — and I'm ready to light this up.",
        lines=3
    )

    voice_desc = gr.Textbox(
        label="Voice Description",
        value=DEFAULT_VOICE,
        lines=2,
        info="Use emotion tags: <angry> <calm> <shout> <laugh> <sad> <sarcastic> etc."
    )

    audio_out = gr.Audio(label="Generated Audio")

    btn = gr.Button("Generate")

    btn.click(fn=generate_tts, inputs=[text, voice_desc], outputs=audio_out)

demo.launch()