Do Audio Language Models Hear and Read Distinctive Features Alike?
Study finds most audio language models do not align heard and read features, except voicing in Qwen2.5-Omni.
Researchers test whether audio language models encode a phonological distinctive feature in the same direction when a phoneme is heard and when it is read. Across 6 models, 7 features, and 15 languages from 11 families, only voicing in the two Qwen2.5-Omni models exceeds a random-pairing reference after multiple-testing correction. In three of six models, voicing keeps one audio direction across 14 languages with enough minimal pairs. Model family, not size, predicts which stream represents a feature, and the random reference varies by a factor of seven between models.
- Six models, seven features, and 15 languages were compared.
- Only Qwen2.5-Omni voicing beat the random-pair reference after correction.
- Model family, not size, predicted which stream encodes a feature.
Full article180 words · extracted from arxiv.org · click to collapse
Audio language models pass speech and text through a single decoder. We ask whether that decoder represents a distinctive feature in the same direction when a phoneme is heard and when it is read. For minimal pairs of phonemes differing in one feature, we take the offset between the two members' mean representations. Averaging those offsets gives a direction for each stream, and we measure the cosine between the two. Because the two streams already agree about arbitrary phoneme pairs, we compare every measure against a reference built from random pairings rather than against zero. We apply this to 6 models, 7 features and 15 languages from 11 families. Only voicing in the two Qwen2.5-Omni models exceeds that reference after correction for multiple testing, and the reference varies by a factor of seven between models. In three of the six models, voicing has one direction in audio across the 14 languages with enough minimal pairs to measure it, and every language pair agrees in two of them. The model family, not the model size, predicts which stream represents a feature.
Text extracted automatically; images, tables and formatting may be missing. Original: https://arxiv.org/abs/2609.30167