Announcement_16_slt

We are proud to announce that Phone Segmentation and Recognition through Phonological Activation Mapping, or SPAM (S3M-based Phonological Activation Mapping) for short, has been accepted to SLT 2026!

Phonetic structure is already present in the representations of self-supervised speech models (S3Ms), and we show that they can be steered to solve phone segmentation and recognition at the same time. The pipeline is simple and intuitive: S3M representations, phonological feature activations, and two gradient-descent-free prediction heads. Our method requires less than a minute of phonetic transcriptions and generalizes well to phones unseen during training!