export_onnx.py

import argparse
from pathlib import Path
from typing import Optional

import torch

import utils
from models import SynthesizerTrn
from text.symbols import symbols

OPSET_VERSION = 15


def main() -> None:
    torch.manual_seed(1234)

    parser = argparse.ArgumentParser()
    parser.add_argument(
        "--model-path", required=True, help="Path to model weights (.pth)"
    )
    parser.add_argument(
        "--config-path", required=True, help="Path to model config (.json)"
    )
    parser.add_argument("--output", required=True, help="Path to output model (.onnx)")

    args = parser.parse_args()

    args.model_path = Path(args.model_path)
    args.config_path = Path(args.config_path)
    args.output = Path(args.output)
    args.output.parent.mkdir(parents=True, exist_ok=True)

    hps = utils.get_hparams_from_file(args.config_path)

    if (
        "use_mel_posterior_encoder" in hps.model.keys()
        and hps.model.use_mel_posterior_encoder == True
    ):
        print("Using mel posterior encoder for VITS2")
        posterior_channels = 80  # vits2
        hps.data.use_mel_posterior_encoder = True
    else:
        print("Using lin posterior encoder for VITS1")
        posterior_channels = hps.data.filter_length // 2 + 1
        hps.data.use_mel_posterior_encoder = False

    model_g = SynthesizerTrn(
        len(symbols),
        posterior_channels,
        hps.train.segment_size // hps.data.hop_length,
        n_speakers=hps.data.n_speakers,
        **hps.model,
    )

    _ = model_g.eval()

    _ = utils.load_checkpoint(args.model_path, model_g, None)

    def infer_forward(text, text_lengths, scales, sid=None):
        noise_scale = scales[0]
        length_scale = scales[1]
        noise_scale_w = scales[2]
        audio = model_g.infer(
            text,
            text_lengths,
            noise_scale=noise_scale,
            length_scale=length_scale,
            noise_scale_w=noise_scale_w,
            sid=sid,
        )[0]

        return audio

    model_g.forward = infer_forward

    dummy_input_length = 50
    sequences = torch.randint(
        low=0, high=len(symbols), size=(1, dummy_input_length), dtype=torch.long
    )
    sequence_lengths = torch.LongTensor([sequences.size(1)])

    sid: Optional[torch.LongTensor] = None
    if hps.data.n_speakers > 1:
        sid = torch.LongTensor([0])

    # noise, length, noise_w
    scales = torch.FloatTensor([0.667, 1.0, 0.8])
    dummy_input = (sequences, sequence_lengths, scales, sid)

    # Export
    torch.onnx.export(
        model=model_g,
        args=dummy_input,
        f=str(args.output),
        verbose=False,
        opset_version=OPSET_VERSION,
        input_names=["input", "input_lengths", "scales", "sid"],
        output_names=["output"],
        dynamic_axes={
            "input": {0: "batch_size", 1: "phonemes"},
            "input_lengths": {0: "batch_size"},
            "output": {0: "batch_size", 1: "time1", 2: "time2"},
        },
    )

    print(f"Exported model to {args.output}")


if __name__ == "__main__":
    main()