Which onnx does the DTLN_norm_500h.h5 file correspond to?

[cpcgskill]

I use DTLN in my project and after trying DTLN_norm_500h.h5 it works best. But I encountered trouble during deployment. Because my project is a desktop software, I cannot carry tf.

[ProtoSol]

How did you used the model??

[cpcgskill]

Currently, I'm using the code directly from the repository and invoking it via the command line. However, since TensorFlow itself is quite large, integrating it causes my tool's size to increase by almost 10 times. Because of that, I’d like to convert the pretrained_model/DTLN_norm_500h.h5 file into ONNX format and use ONNX for inference instead.

Below is my complete invocation code:

import shutil

if False:
    from typing import *
import os
import sys

from command import call_command
from processor._utils import _get_sub_dir_path

eval_path = 'bin/DTLN/run_evaluation.py'
model_path = 'bin/DTLN/DTLN_norm_500h.h5'

if not os.path.isfile(eval_path) or not os.path.isfile(model_path):
    sys.exit(1)


def DTLN_batch_process(input_files):
    in_tmp_dir = _get_sub_dir_path()
    out_tmp_dir = _get_sub_dir_path()
    in_files = []
    for i in input_files:
        if os.path.isfile(i):
            in_files.append(shutil.copy(i, in_tmp_dir))
    call_command([sys.executable, eval_path, '-i', in_tmp_dir, '-o', out_tmp_dir, '-m', model_path])
    return [os.path.join(out_tmp_dir, os.path.basename(i)) for i in in_files]