What all do audio transformer models hear? Probing Acoustic Representations for Language Delivery and its Structure
This research helps understand the internal workings of audio transformer models, which is important for researchers and developers aiming to optimize their use in speech processing tasks.
This paper investigates what information audio transformer models like Mockingjay and wave2vec2.0 learn by probing their acoustic representations for language delivery and structure features. The study compares these models on audio, fluency, and pronunciation features, as well as textual surface, syntax, and semantic features, across diverse speech datasets.
In recent times, BERT based transformer models have become an inseparable part of the 'tech stack' of text processing models. Similar progress is being observed in the speech domain with a multitude of models observing state-of-the-art results by using audio transformer models to encode speech. This begs the question of what are these audio transformer models learning. Moreover, although the standard methodology is to choose the last layer embedding for any downstream task, but is it the optimal choice? We try to answer these questions for the two recent audio transformer models, Mockingjay and wave2vec2.0. We compare them on a comprehensive set of language delivery and structure features including audio, fluency and pronunciation features. Additionally, we probe the audio models' understanding of textual surface, syntax, and semantic features and compare them to BERT. We do this over exhaustive settings for native, non-native, synthetic, read and spontaneous speech datasets