Scaling Properties of Speech Language Models

Anonymous

Abstract: Speech Language Models (SLMs) aim to learn language from raw audio, without textual resources. Despite significant advances, our current models exhibit weak syntax and semantic abilities. However, if the scaling properties of neural language models hold for the speech modality, these abilities will improve as the amount of compute used for training increases. In this paper, we use models of this scaling behavior to estimate the scale at which our current methods will yield a SLM with the English proficiency of text-based Large Language Models (LLMs). We establish a strong correlation between pre-training loss and downstream syntactic and semantic performance in SLMs and LLMs, which results in predictable scaling of linguistic performance. We show that the linguistic performance of SLMs scales up to three orders of magnitude more slowly than the performance of text-based LLMs. Additionally, we study the effects of coarser speech tokenization, and the benefits of synthetic data designed to boost semantic understanding.

Paper Type: long

Research Area: Speech recognition, text-to-speech and spoken language understanding

Contribution Types: Model analysis & interpretability, NLP engineering experiment, Data resources

Languages Studied: English