findsylls: A Language-Agnostic Toolkit for Syllable-Level Speech Tokenization and Embedding

Dimensions overall score 5.0

GitHub Code Pulse

No public code linked for this paper yet.

• findsylls is a modular, language-agnostic toolkit that unifies classical syllable detectors and end-to-end syllabifiers under a common interface.method 95%
• findsylls enables controlled comparisons of representations, algorithms, and token rates by allowing their components to be recombined.method 95%
• findsylls supports reproducible syllable-level experiments across both high-resource and under-resourced settings.method 95%
• Syllable-level tokenization can match or surpass high-frame-rate SSL tokens on spoken language understanding benchmarks while reducing training time more than twofold and FLOPs by approximately fivefold.result 90%
• Mismatches between a SSL model’s pretraining languages and the target language can reduce segmentation accuracy by 20–30%.result 85%
• All tested approaches produce only a few syllabic tokens per second of audio (3.1–5.8 tok/s), representing a large reduction relative to typical 50–100 Hz frame-level features.result 90%
• findsylls allows direct tests of how different segmentation algorithms behave with alternative encoders by enabling the use of BERT or VG-HuBERT features.method 90%
• The toolkit supports evaluation on Kono, an underdocumented Central Mande language, using newly annotated data.method 95%

Competitive landscape

Competitor map is still being generated for this paper. Enable generation or check back soon.

Startup potential card