Pith. sign in

Distilling a speech and music encoder with task arithmetic

1 Pith paper cite this work. Polarity classification is still indexing.

1 Pith paper citing it
abstract

Despite the progress in self-supervised learning (SSL) for speech and music, existing models treat these domains separately, limiting their capacity for unified audio understanding. A unified model is desirable for applications that require general representations, e.g. audio large language models. Nonetheless, directly training a general model for speech and music is computationally expensive. Knowledge Distillation of teacher ensembles may be a natural solution, but we posit that decoupling the distillation of the speech and music SSL models allows for more flexibility. Thus, we propose to learn distilled task vectors and then linearly interpolate them to form a unified speech+music model. This strategy enables flexible domain emphasis through adjustable weights and is also simpler to train. Experiments on speech and music benchmarks demonstrate that our method yields superior overall performance compared to ensemble distillation.

fields

eess.AS 1

years

2025 1

verdicts

CONDITIONAL 1

representative citing papers

Multi-Distillation from Speech and Music Representation Models

eess.AS · 2025-06-08 · conditional · novelty 4.0

A 23M-parameter student distilled from HuBERT/WavLM and MERT gets close to teacher-level average accuracy on speech and music benchmarks and outperforms its teachers in few-shot classification.

citing papers explorer

Showing 1 of 1 citing paper.

  • Multi-Distillation from Speech and Music Representation Models eess.AS · 2025-06-08 · conditional · none · ref 13 · internal anchor

    A 23M-parameter student distilled from HuBERT/WavLM and MERT gets close to teacher-level average accuracy on speech and music benchmarks and outperforms its teachers in few-shot classification.