Bittner, M., Schnöll, D., Dallinger, D., Wess, M., & Jantsch, A. (2025). Pruning State Space Models with Model Order Reduction for Efficient Raw Audio Classification. In 33rd European Signal Processing Conference EUSIPCO 2025 (pp. 271–275). IEEE. http://hdl.handle.net/20.500.12708/226014
E384-02 - Forschungsbereich Systems on Chip E056-10 - Fachbereich SecInt-Secure and Intelligent Human-Centric Digital Technologies E056-16 - Fachbereich SafeSeclab
-
Erschienen in:
33rd European Signal Processing Conference EUSIPCO 2025
-
ISBN:
978-9-46-459362-4
-
Datum (veröffentlicht):
2025
-
Veranstaltungsname:
33rd European Signal Processing Conference EUSIPCO 2025
en
Veranstaltungszeitraum:
8-Sep-2025 - 12-Sep-2025
-
Veranstaltungsort:
Palermo, Italien
-
Umfang:
5
-
Verlag:
IEEE
-
Peer Reviewed:
Ja
-
Keywords:
Pruning; Model Order Reduction; Deep State Space Models; Raw Audio Classification
en
Abstract:
Deep State Space Models (SSMs) have shown good performance on long-sequence classification tasks such as raw
audio classification. Targeting edge devices it is crucial to further improve their inference efficiency. However, pruning techniques
are not well explored for SSMs. We propose a layer-wise Model Order Reduction (MOR) technique based on balanced truncation combined with an iterative pruning algorithm to increase the efficiency of already trained SSM models, without the need for retraining. Specifically, we focus on S-Edge models, a class of hardware-friendly SSMs. Evaluated on the Google Speech Commands dataset we prune models ranging from 141k–8k in parameters and 94.9%–90.0% in test accuracy. Given an
accuracy loss constraint of 0.5pp we are able to find models which reduce parameters by 36.1% for the biggest and 5.8% for
the smallest model.