app.py

import gradio as gr
from transformers import VitsModel, AutoTokenizer
import torch
import scipy.io.wavfile
import numpy as np
import librosa
import soundfile as sf
import tempfile

# Load the model and tokenizer
model = VitsModel.from_pretrained("facebook/mms-tts-eng")
tokenizer = AutoTokenizer.from_pretrained("facebook/mms-tts-eng")


def pitch_shift_np(audio_np, sampling_rate, pitch_shift):
    # Correcting the function call
    return librosa.effects.pitch_shift(audio_np, sr=sampling_rate, n_steps=pitch_shift)


def synthesize_speech(text, pitch_shift):
    # Tokenize the input text
    inputs = tokenizer(text, return_tensors="pt")

    # Generate waveform
    with torch.no_grad():
        output = model(**inputs).waveform.squeeze().numpy()

    # Pitch shift
    shifted_audio = pitch_shift_np(output, model.config.sampling_rate, pitch_shift)

    # Save to a temporary file
    with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as fp:
        sf.write(fp.name, shifted_audio, model.config.sampling_rate)
        temp_file_path = fp.name

    return temp_file_path


# Create the Gradio interface
interface = gr.Interface(
    fn=synthesize_speech,
    inputs=[
        gr.components.Textbox(lines=2, placeholder="Type your text here..."),
        gr.components.Slider(minimum=-2, maximum=2, step=0.1, label="Pitch Shift (Semitones)")
    ],
    outputs=gr.components.Audio(type="filepath", label="Generated Speech"),
    title="Text to Speech Synthesis",
    description="Type text and convert it to speech using a TTS model. Use the slider to adjust the pitch."
)

# Launch the application
interface.launch()