Scaling Spoken Language Models with Syllabic Speech Tokenization (2509.26634v1)

Abstract: Spoken LLMs (SLMs) typically discretize speech into high-frame-rate tokens extracted from SSL speech models. As the most successful LMs are based on the Transformer architecture, processing these long token streams with self-attention is expensive, as attention scales quadratically with sequence length. A recent SSL work introduces acoustic tokenization of speech at the syllable level, which is more interpretable and potentially more scalable with significant compression in token lengths (4-5 Hz). Yet, their value for spoken LLMing is not yet fully explored. We present the first systematic study of syllabic tokenization for spoken LLMing, evaluating models on a suite of SLU benchmarks while varying training data scale. Syllabic tokens can match or surpass the previous high-frame rate tokens while significantly cutting training and inference costs, achieving more than a 2x reduction in training time and a 5x reduction in FLOPs. Our findings highlight syllable-level LLMing as a promising path to efficient long-context spoken LLMs.