Performance Analysis of Speech Encoders for Low-Resource SLU and ASR in Tunisian Dialect
Salima Mdhaffar, Haroun Elleuch, Fethi Bougares, Yannick Estève
TL;DR
The paper tackles evaluating SSL speech encoders for ASR and SLU in a low-resource Tunisian Arabic dialect using the TARIC-SLU dataset. It benchmarks 14 SSL models spanning monolingual, cross-lingual, and teacher-student refined variants, plus Whisper baselines, under a unified protocol. Key findings show w2v-BERT 2.0 (600M parameters, trained on 4.5M hours across 143 languages) often yields the best ASR and SLU performance, while wavLM excels among monolingual encoders for ASR, SAMU-XLSR benefits from semantic refinement, and data2vec 2.0 excels on semantically rich utterances. The study provides actionable guidance for deploying SLU/ASR in very low-resource dialects and motivates further development of hybrid semantic-transfer models.
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.