Preprint · arXiv · September 9, 2026
Orukeet: Multilingual ASR with Frozen Gabor Kernels
Nathan Roll, Irene Yi, Büşra Marşan, Vianney Grenez, Gabriel Stein, Momcilo Mrkaic
Abstract
Orukeet replaces half of an adapted Parakeet encoder's temporal filters with 12,288 fitted Gabor kernels, freezes these replacements, and trains the remaining parameters on multilingual and multi-accent data. Final adaptation and checkpoint selection use LibriSpeech test-other. Across 20,146 FLEURS recordings in 25 languages, pooled word error rate (WER) falls from Parakeet's 11.01% to Orukeet's 9.85%, a 10.6% relative reduction. Orukeet has lower WER on 23 of the 25 languages. Orukeet outperforms Parakeet on 61 out of 74 tested splits, including LibriSpeech test-clean (1.46% vs. 1.53% WER), test-other (2.86% vs. 3.14%), and FLEURS English (3.82% vs. 4.28%). All comparisons decode the same audio with matched NeMo settings. The fitted kernels are stored as ordinary convolution weights, retaining Parakeet's architecture and inference operators.
More by Büşra Marşan
- Translanguaging Pedagogy in the Age of AIContemporary research in language and linguistics. · 2026 · Citations: 1
- Universal Dependencies 2.17Lindat/Clarin Repository · 2025
- Strategies for the Annotation of Pronominalised Locatives in Turkic Universal Dependency TreebanksLanguage Resources and Evaluation Conference · 2024 · Citations: 1
- Evaluating the Quality of a Corpus Annotation Scheme Using Pretrained Language ModelsLanguage Resources and Evaluation Conference · 2024
