InferenceInterfaces/ControllableInterface.py

import os

import torch

from InferenceInterfaces.ToucanTTSInterface import ToucanTTSInterface
from Modules.ControllabilityGAN.GAN import GanWrapper
from Utility.storage_config import MODELS_DIR


class ControllableInterface:

    def __init__(self, gpu_id="cpu", available_artificial_voices=1000):
        if gpu_id == "cpu":
            os.environ["CUDA_VISIBLE_DEVICES"] = ""
        else:
            os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"
            os.environ["CUDA_VISIBLE_DEVICES"] = f"{gpu_id}"
        self.device = "cuda" if gpu_id != "cpu" else "cpu"
        self.model = ToucanTTSInterface(device=self.device, tts_model_path=None)
        self.model.set_utterance_embedding("female_high_voice.wav")

    def read(self,
             prompt,
             loudness_in_db
             ):
        phones = self.model.text2phone.get_phone_string(prompt)
        print(prompt + "\n" + phones + "\n\n")

        if len(phones) > 1800:
            prompt = "Your input was too long. Please try either a shorter text or split it into several parts."

        wav, sr = self.model(phones,
                                  input_is_phones=True,
                                  duration_scaling_factor=1.0,
                                  pitch_variance_scale=1.0,
                                  energy_variance_scale=1.0,
                                  pause_duration_scaling_factor=1.0,
                                  return_plot_as_filepath=False,
                                  prosody_creativity=0.8,
                                  loudness_in_db=loudness_in_db)
        return sr, wav