On the reliability of feature attribution methods for speech classification

TL;DR

This paper investigates the reliability of feature attribution methods in speech classification, revealing that most standard approaches are unreliable in speech tasks except for word-aligned perturbation methods in word-based classifications.

Contribution

The study systematically evaluates how input type, aggregation, and perturbation timespan affect feature attribution reliability in speech models, highlighting domain-specific challenges.

Findings

Standard feature attribution methods are generally unreliable for speech.

Word-aligned perturbation methods work well for word-based classification tasks.

Reliability depends on input characteristics and task specifics.

Abstract

As the capabilities of large-scale pre-trained models evolve, understanding the determinants of their outputs becomes more important. Feature attribution aims to reveal which parts of the input elements contribute the most to model outputs. In speech processing, the unique characteristics of the input signal make the application of feature attribution methods challenging. We study how factors such as input type and aggregation and perturbation timespan impact the reliability of standard feature attribution methods, and how these factors interact with characteristics of each classification task. We find that standard approaches to feature attribution are generally unreliable when applied to the speech domain, with the exception of word-aligned perturbation methods when applied to word-based classification tasks.