Performance Analysis of Speech Encoders for Low-Resource SLU and ASR in Tunisian Dialect

TL;DR

This paper evaluates various self-supervised speech encoders for low-resource Tunisian Arabic SLU and ASR, highlighting their effectiveness and limitations in a challenging low-data scenario.

Contribution

It compares SSL speech encoders in low-resource Tunisian dialect settings, including monolingual, multilingual, and teacher-student refined models, for SLU and ASR tasks.

Findings

SSL models improve low-resource SLU and ASR performance

Multilingual and teacher-student models outperform monolingual ones

Limited Tunisian data challenges SSL model fine-tuning

Abstract

Speech encoders pretrained through self-supervised learning (SSL) have demonstrated remarkable performance in various downstream tasks, including Spoken Language Understanding (SLU) and Automatic Speech Recognition (ASR). For instance, fine-tuning SSL models for such tasks has shown significant potential, leading to improvements in the SOTA performance across challenging datasets. In contrast to existing research, this paper contributes by comparing the effectiveness of SSL approaches in the context of (i) the low-resource spoken Tunisian Arabic dialect and (ii) its combination with a low-resource SLU and ASR scenario, where only a few semantic annotations are available for fine-tuning. We conduct experiments using many SSL speech encoders on the TARIC-SLU dataset. We use speech encoders that were pre-trained on either monolingual or multilingual speech data. Some of them have also been refined without in-domain nor Tunisian data through multimodal supervised teacher-student paradigm. This study yields numerous significant findings that we are discussing in this paper.