app.py

import gradio as gr
import torch
import librosa
from pathlib import Path
import tempfile, torchaudio


# Load the MARS5 model
mars5, config_class = torch.hub.load('Camb-ai/mars5-tts', 'mars5_english', trust_repo=True)

# Default reference audio and transcript
# default_audio_path = "example.wav"
# default_transcript = "We actually haven't managed to meet demand."

# Function to process the text and audio input and generate the synthesized output
def synthesize(text, audio_file, transcript):
    # Load the reference audio
    wav, sr = librosa.load(audio_file, sr=mars5.sr, mono=True)
    wav = torch.from_numpy(wav)
    
    # Define the configuration for the TTS model
    deep_clone = True
    cfg = config_class(deep_clone=deep_clone, rep_penalty_window=100, top_k=100, temperature=0.7, freq_penalty=3)
    
    # Generate the synthesized audio
    ar_codes, wav_out = mars5.tts(text, wav, transcript, cfg=cfg)
    
    # Save the synthesized audio to a temporary file
    output_path = Path(tempfile.mktemp(suffix=".wav"))
    torchaudio.save(output_path, wav_out.unsqueeze(0), mars5.sr)
    
    return str(output_path)

# Create the Gradio interface
interface = gr.Interface(
    fn=synthesize,
    inputs=[
        gr.Textbox(label="Text to synthesize"),
        gr.Audio(label="Audio file to clone from", type="filepath"),
        gr.Textbox(label="Uploaded audio file transcript"),
    ],
    outputs=gr.Audio(label="Synthesized Audio"),
    title="MARS5 TTS Demo",
    description="Enter text and upload an audio file to clone the voice and generate synthesized speech using MARS5 TTS."
)

# Launch the Gradio app
interface.launch()