Spoken Language Modeling with Duration-Penalized Self-Supervised Units
This work addresses the optimization of acoustic units for spoken language models, which is incremental as it explores an underexplored interaction to guide task-specific unit selection.
The paper investigates how codebook size and unit coarseness affect spoken language model performance, finding that coarser units improve sentence-level resynthesis and language modeling tasks at lower bitrates, while phone and word levels benefit more from appropriate codebook size.
Spoken language models (SLMs) operate on acoustic units obtained by discretizing self-supervised speech representations. Although the characteristics of these units directly affect performance, the interaction between codebook size and unit coarseness (i.e., duration) remains unexplored. We investigate SLM performance as we vary codebook size and unit coarseness using the simple duration-penalized dynamic programming (DPDP) method. New analyses are performed across different linguistic levels. At the phone and word levels, coarseness provides little benefit, as long as the codebook size is chosen appropriately. However, when producing whole sentences in a resynthesis task, SLMs perform better with coarser units. In lexical and syntactic language modeling tasks, coarser units also give higher accuracies at lower bitrates. We therefore show that coarser units aren't always better, but that DPDP is a simple and efficient way to obtain coarser units for the tasks where they are beneficial.