Adds multi-language support for VITS onnx, fixes onnx exporting and inference errors

[SystemPanic]

• Adds multi-language support for VITS onnx
• Fixes [Bug] RuntimeError: Given groups=1, weight of size [196, 196, 1], expected input[1, 192, 100] to have 196 channels, but got 192 channels instead #2753 incorrect number of channels when having language_emb_dim > 0
• Fixes onnx inference error when speaker_id is None or not passed as an argument
• Fixes onnx exporting error AttributeError: 'Vits' object has no attribute 'disc' for models with init_discriminator=false

Tested with multi-speaker and multi-language models, and with single speaker and single language, using the following script:

import torch
import os
import numpy as np
from TTS.tts.models.vits import Vits
from TTS.tts.configs.vits_config import VitsConfig
from TTS.utils.audio.numpy_transforms import save_wav

modelPath = "MULTILANG_MULTISPEAKER_PATH"
speaker_id = 0 '''None if no multi-speaker model'''
language_id = 0 '''None if no multi-language model'''

config = VitsConfig()
config.load_json(os.path.join(modelPath, "config.json"))
vits = Vits.init_from_config(config)

vits.load_onnx(os.path.join(modelPath, "MULTILANG_MULTISPEAKER_PATH.onnx"))

text = "LONG TEXT HERE"
text_inputs = np.asarray(
    vits.tokenizer.text_to_ids(text),
    dtype=np.int64,
)[None, :]

audio = vits.inference_onnx(text_inputs, speaker_id=speaker_id, language_id=language_id)
save_wav(wav=audio[0], path=os.path.join(os.path.dirname(__file__), 'test.wav'), sample_rate=config.audio.sample_rate)

[CLAassistant]

All committers have signed the CLA.

[erogol]

Thanks for the PR. There is one CI error but it is not about your PR.