app.py

import gradio as gr
import torch
from transformers import AutoProcessor, AutoModelForTextToWaveform
import scipy.io.wavfile as wavfile
import numpy as np
import tempfile

# -----------------------------
# Load Bark Model + Processor
# -----------------------------
device = "cuda" if torch.cuda.is_available() else "cpu"

processor = AutoProcessor.from_pretrained("suno/bark")

model = AutoModelForTextToWaveform.from_pretrained(
    "suno/bark",
    torch_dtype=torch.float16 if device == "cuda" else torch.float32,
).to(device)


# -----------------------------
# Voice Presets (Bark v2 manual list)
# -----------------------------
voices = [
    "v2/en_speaker_0",
    "v2/en_speaker_1",
    "v2/en_speaker_2",
    "v2/en_speaker_3",
    "v2/en_speaker_4",
    "v2/en_speaker_5",
    "v2/en_speaker_6",
    "v2/en_speaker_7",
    "v2/en_speaker_8",
    "v2/en_speaker_9",
]

voices = sorted(voices)



# -----------------------------
# Audio Generation Function
# -----------------------------
def generate_audio(text, voice):
    if not text.strip():
        return None

    # Prepare inputs
    inputs = processor(text, voice_preset=voice)

    # Move tensors to model.device
    for k, v in inputs.items():
        if torch.is_tensor(v):
            inputs[k] = v.to(device)

    # Generate waveform
    with torch.no_grad():
        audio = model.generate(**inputs)

    # Convert to numpy
    audio = audio.cpu().numpy().squeeze()
    sample_rate = 24000

    # Save temporary WAV file for Gradio output
    with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as fp:
        wavfile.write(fp.name, sample_rate, audio)
        return fp.name


# -----------------------------
# Gradio Interface
# -----------------------------
demo = gr.Interface(
    fn=generate_audio,
    inputs=[
        gr.Textbox(
            label="Prompt",
            placeholder="Type what you want Bark to say or sing...",
            lines=3
        ),
        gr.Dropdown(
            voices,
            label="Voice Preset",
            value="v2/en_speaker_6"
        ),
    ],
    outputs=gr.Audio(label="Generated Audio"),
    title="🎤 Bark Text-to-Audio (Suno, via HuggingFace Transformers)",
    description="Generates speech, singing, music, and sound effects using the open-source Bark model.",
)

# -----------------------------
# Launch
# -----------------------------
demo.launch()