app.py

import os
import torch
import gradio as gr
import torchaudio
import time
from datetime import datetime
from tortoise.api import TextToSpeech
from tortoise.utils.text import split_and_recombine_text
from tortoise.utils.audio import load_audio, load_voice, load_voices

VOICE_OPTIONS = [
    "kasuri",
    "shanuka",
    "indian_f_1",
]

def inference(
    text,
    voice
):
    if text is None or text.strip() == "":
        raise gr.Error("Please provide text.")

    texts = split_and_recombine_text(text)

    voice_samples, conditioning_latents = load_voice(voice)

    start_time = time.time()

    for j, text in enumerate(texts):
        for audio_frame in tts.tts_with_preset(
            text,
            voice_samples=voice_samples,
            conditioning_latents=conditioning_latents,
            preset="ultra_fast",
            k=1
        ):
            yield (24000, audio_frame.cpu().detach().numpy())

def main():
    title = "Tortoise TTS"
    description = """
    """
    text = gr.Textbox(
        lines=4,
        label="Text:",
    )

    voice = gr.Dropdown(
        VOICE_OPTIONS, value="kasuri", label="Select voice:", type="value"
    )

    output_audio = gr.Audio(label="streaming audio:", streaming=True, autoplay=True)

    interface = gr.Interface(
        fn=inference,
        inputs=[
            text,
            voice
        ],
        title=title,
        description=description,
        outputs=[output_audio],
    )
    interface.queue().launch()

if __name__ == "__main__":
    tts = TextToSpeech(kv_cache=True, use_deepspeed=True, half=True)

    with open("Tortoise_TTS_Runs_Scripts.log", "a") as f:
        f.write(
            f"\n\n-------------------------Tortoise TTS Scripts Logs, {datetime.now()}-------------------------\n"
        )

    main()