app/speech.py

import torch
from TTS.api import TTS
from loguru import logger

class Speech():
    
    def __init__(self) -> None:
        
        # List available 🐸TTS models
        # print(TTS().list_models())

        # Get device
        device = "cuda" if torch.cuda.is_available() else "cpu"

        # Init TTS (and download model if necessary)
        self.tts = TTS("tts_models/multilingual/multi-dataset/xtts_v1").to(device)

        logger.info(f"Initialized TTS: {self.tts}")

    def text_to_speech(self, text:str, language:str):

        logger.info(f"Text to speech: {text} to {language}")

        # remove ## from text
        text = text.replace("##", "")

        # language =  en, de, es, fr, it, pt, pl.
        valid_languages = ["en", "de", "es", "fr", "it", "pt", "pl"]
        if language not in valid_languages:
            raise ValueError(f"Language '{language}' is not one of the supported languages: {valid_languages}")
            return None 
        else:
            # Run TTS
            # ❗ Since this model is multi-lingual voice cloning model, we must set the target speaker_wav and language
            # Text to speech list of amplitude values as output
            # wav = self.tts.tts(text=text, language=language)
            # # Text to speech to a file
            self.tts.tts_to_file(text=text, language=language, file_path="output.wav")

            logger.info(f"Converted text to speech")
            return "output.wav"