app.py

import os
import torch
import gradio as gr
import torchaudio
import time
import spaces
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

tts = TextToSpeech(kv_cache=True)

@spaces.GPU
def inference(
    text,
    reference_audio,
):
    output_wav_path = tempfile.mktemp(suffix=".wav")
    texts = split_and_recombine_text(text)

    start_time = time.time()

    all_parts = []
    for j, text in enumerate(texts):
        for audio_frame in tts.tts_with_preset(
            text,
            voice_samples=load_audio(init_audio_file),
            preset="fast",
        ):
            # print("Time taken: ", time.time() - start_time)
            all_parts.append(audio_frame)
            # yield (24000, audio_frame.cpu().detach().numpy())

    wav = torch.cat(all_parts, dim=0).unsqueeze(0)
    print(wav.shape)
    torchaudio.save(output_wav_path, wav.cpu(), 24000)

    return output_wav_path
    
def main():
    title = "Tortoise TTS 🐢"

    text = gr.Textbox(
        label="Text",
    )

    reference_audio = gr.Audio(label="Reference Audio", type="filepath")

    output_audio = gr.Audio(label="Generated Speech")
    # download_audio = gr.Audio(label="dowanload audio:")
    interface = gr.Interface(
        fn=inference,
        inputs=[
            text,
            reference_audio,
        ],
        title=title,
        outputs=output_audio,
    )
    interface.launch()


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

    main()