app.py

import gradio as gr
import torch
import os
from io import BytesIO
import base64
import numpy as np
from pydub import AudioSegment
from parler_tts import ParlerTTSForConditionalGeneration
from transformers import AutoTokenizer, AutoFeatureExtractor, set_seed

device = "cuda:0" if torch.cuda.is_available() else "cpu"

# Load the TTS model from Huggingface Hub
repo_id = "parler-tts/parler_tts_mini_v0.1"
model = ParlerTTSForConditionalGeneration.from_pretrained(repo_id).to(device)
tokenizer = AutoTokenizer.from_pretrained(repo_id)
feature_extractor = AutoFeatureExtractor.from_pretrained(repo_id)

SAMPLE_RATE = feature_extractor.sampling_rate
SEED = 42

# Secret token for authenticating requests
SECRET_TOKEN = os.getenv('SECRET_TOKEN', 'default_secret')

def generate_audio(secret_token, text, description):
    # Token-based access
    if secret_token != SECRET_TOKEN:
        raise gr.Error('Invalid secret token.')

    # Tokenize input text and description
    inputs = tokenizer(description, return_tensors="pt").to(device)
    prompt = tokenizer(text, return_tensors="pt").to(device)

    # Set the random seed for reproducibility
    set_seed(SEED)

    # Generate speech from text
    generation = model.generate(
        input_ids=inputs.input_ids,
        prompt_input_ids=prompt.input_ids,
        do_sample=True,
        temperature=1.0
    )

    # Convert output to numpy array
    audio_arr = generation.cpu().numpy().squeeze()
    samples = np.array(audio_arr * (2**15 - 1), dtype=np.int16)

    # Create audio segment from numpy data
    sound = AudioSegment(
        samples.tobytes(),
        frame_rate=SAMPLE_RATE,
        sample_width=samples.dtype.itemsize,
        channels=1
    )

    # Export the audio to MP3 format
    buff_mp3 = BytesIO()
    sound.export(buff_mp3, format="mp3")
    buff_mp3.seek(0)

    # Encode the MP3 to base64
    audio_base64 = base64.b64encode(buff_mp3.read()).decode('utf-8')
    audio_uri = 'data:audio/mp3;base64,' + audio_base64

    return audio_uri

# Gradio interface
with gr.Blocks() as app:
    gr.HTML("""
        <div style="text-align: center;">
            <h3>TTS Audio Generator</h3>
        </div>
    """)

    secret_token = gr.Textbox(label="Secret Token")
    input_text = gr.Textbox(label="Text Input")
    description = gr.Textbox(label="Description")
    run_button = gr.Button("Generate Audio")
    audio_output = gr.Audio(label="Generated Audio", type="auto")

    inputs = [secret_token, input_text, description]
    outputs = [audio_output]

    run_button.click(fn=generate_audio, inputs=inputs, outputs=outputs, queue=True)

app.queue()
app.launch()