Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

AraSpot: Arabic Spoken Command Spotting

Mahmoud Salhab, Haidar Harmanani

TL;DR

Arabic spoken keyword spotting on edge devices faces data scarcity and the need for efficient, accurate models. The paper introduces AraSpot, integrating online data augmentation, Tacotron 2–based synthetic data with WaveGlow, and a ConformerGRU architecture to capture both local and long-range speech patterns. It achieves state-of-the-art performance on the ASC Arabic dataset with 40 keywords, attaining 99.59% accuracy, and demonstrates substantial gains over prior Arabic KWS methods. The approach offers a practical pathway for robust, low-resource Arabic voice activation in real-time devices. The work highlights the value of combining synthetic data with advanced hybrid architectures for improved robustness in spoken language understanding tasks.

Abstract

Spoken keyword spotting (KWS) is the task of identifying a keyword in an audio stream and is widely used in smart devices at the edge in order to activate voice assistants and perform hands-free tasks. The task is daunting as there is a need, on the one hand, to achieve high accuracy while at the same time ensuring that such systems continue to run efficiently on low power and possibly limited computational capabilities devices. This work presents AraSpot for Arabic keyword spotting trained on 40 Arabic keywords, using different online data augmentation, and introducing ConformerGRU model architecture. Finally, we further improve the performance of the model by training a text-to-speech model for synthetic data generation. AraSpot achieved a State-of-the-Art SOTA 99.59% result outperforming previous approaches.

AraSpot: Arabic Spoken Command Spotting

TL;DR

Arabic spoken keyword spotting on edge devices faces data scarcity and the need for efficient, accurate models. The paper introduces AraSpot, integrating online data augmentation, Tacotron 2–based synthetic data with WaveGlow, and a ConformerGRU architecture to capture both local and long-range speech patterns. It achieves state-of-the-art performance on the ASC Arabic dataset with 40 keywords, attaining 99.59% accuracy, and demonstrates substantial gains over prior Arabic KWS methods. The approach offers a practical pathway for robust, low-resource Arabic voice activation in real-time devices. The work highlights the value of combining synthetic data with advanced hybrid architectures for improved robustness in spoken language understanding tasks.

Abstract

Spoken keyword spotting (KWS) is the task of identifying a keyword in an audio stream and is widely used in smart devices at the edge in order to activate voice assistants and perform hands-free tasks. The task is daunting as there is a need, on the one hand, to achieve high accuracy while at the same time ensuring that such systems continue to run efficiently on low power and possibly limited computational capabilities devices. This work presents AraSpot for Arabic keyword spotting trained on 40 Arabic keywords, using different online data augmentation, and introducing ConformerGRU model architecture. Finally, we further improve the performance of the model by training a text-to-speech model for synthetic data generation. AraSpot achieved a State-of-the-Art SOTA 99.59% result outperforming previous approaches.
Paper Structure (11 sections, 6 equations, 2 figures, 3 tables)

This paper contains 11 sections, 6 equations, 2 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Dataset Description
  4. Solution Approach
  5. Data Augmentation
  6. Synthetic Data Generation Using TTS
  7. ConformerGRU Model
  8. Experiments and results
  9. Experiments Setup
  10. Results
  11. Conclusion and future work

Figures (2)

  • Figure 1: ConformerGRU model architecture
  • Figure 2: Analysis of AraSpot performance under various scenarios, illustrating the model parameters (dimensionality, number of heads, and layers) on the X-axis and corresponding accuracy on the Y-axis. The horizontal black line represents the accuracy of the optimal model from the literature asc. Results are presented for models trained on original data with synthetic data generated through TTS and online data augmentation (depicted by blue bars), as well as models trained solely on original data with online data augmentation (depicted by orange bars).