Data preparation instructions about vq-mae-s-code HOT 2 CLOSED

@samsad35 Okay awesome. Thanks for the prompt and detailed instructions. I will let you know if I have any additional questions in preparing the data. I look forward to the explanations in the future! :)

from vq-mae-s-code.

@willxxy Thank you for your interest in our work. Regarding the pre-training process, we trained a basic VQVAE on audio speech (power spectrogram). We then saved the discrete indices generated by the VQVAE on the VoxCeleb database into an H5 file. In the "VoxcelebSequential" dataset, we directly read the H5 file to retrieve the discrete indices, bypassing the need for extensive preprocessing steps such as reading the audio file, generating the spectrogram, VQVAE encoding, and quantization. We appreciate your feedback, and we will include a dedicated section to explain these steps and provide clear instructions.

from vq-mae-s-code.