What is embedding_model.onnx and melspectrogram.onnx?

[sangheonEN]

When I created the F object below, I used the embedding_model.onnx and melspectrogram.onnx models provided by openwakeword.

F = openwakeword.utils.AudioFeatures()

These are not models developed for my dataset, but models provided by openwakeword, so could the performance not be optimized during training?

If so, could you tell me how to create the embedding_model.onnx and melspectrogram.onnx models that fit my dataset?

Code content. class AudioFeatures():
"""
A class for creating audio features from audio data, including melspectograms and Google's
speech_embedding features.
"""
def init(self,
melspec_model_path: str = "",
embedding_model_path: str = "",
sr: int = 16000;
ncpu: int = 1;
inference_framework: str = "onnx",
device: str = 'cpu'
):
"""
Initialize the AudioFeatures object.

Args:
melspec_model_path (str): The path to the model for computing melspectograms from audio data
embedding_model_path (str): The path to the model for Google's speech_embedding model
sr (int): The sample rate of the audio (default: 16000 khz)