On the Inductive Bias of Word-Character-Level Multi-Task Learning for Speech RecognitionDownload PDF

Anonymous

Published: 16 Nov 2018, Last Modified: 05 May 2023NIPS 2018 Workshop IRASL Blind SubmissionReaders: Everyone
Abstract: End-to-end automatic speech recognition (ASR) commonly transcribes audio signals into sequences of characters while its performance is evaluated by measuring the word-error rate (WER). This suggests that predicting sequences of words directly may be helpful instead. However, training with word-level supervision can be more difficult due to the sparsity of examples per label class. In this paper we analyze an end-to-end ASR model that combines a word-and-character representation in a multi-task learning (MTL) framework. We show that it improves on the WER and study how the word-level model can benefit from character-level supervision by analyzing the learned inductive preference bias of each model component empirically. We find that by adding character-level supervision, the MTL model interpolates between recognizing more frequent words (preferred by the word-level model) and shorter words (preferred by the character-level model).
Keywords: speech recognition, multi-task learning, interpretability
TL;DR: Multi-task learning improves word-and-character-level speech recognition by interpolating the preference biases of its components: frequency- and word length-preference.
7 Replies

Loading