Back to AIBriefs
AnalysisAI Models

Wavelet as Tokenizer: Shared token schema for audio, images, video

Arxiv CS.SD (Sound/Audio ML)

The paper introduces a preliminary continuous-token model using a one-level Haar DWT for audio, images, and video. It aims to replace separate modality-specific latent grids with a shared wavelet token schema. Results are preliminary.

·
8 days ago
Wavelet as Tokenizer: Shared token schema for audio, images, video — AIBriefs