USING EMBEDDED MACHINE LEARNING IN THE PHYSICAL WORLD TO DETECT TOXICITY IN SPOKEN LANGUAGE

Abstract

Toxicity is a prevalent social behavior that involves the use of hate speech, offensive language, bullying, and abusive speech. While text-based approaches for toxicity detection are common, there is limited research on processing speech signals in the physical world. Detecting toxicity in the physical world is challenging due to the difficulty of integrating AI-capable computers into the environment. We propose a lightweight transformer model based on wav2vec2.0 and optimize it using quantization and knowledge distillation techniques. Our model uses multitask learning and achieves an average macro F1-score of 90.3% and a weighted accuracy of 88%, outperforming state-ofthe- art methods on DeToxy-B and the IEMOCAP datasets. In our study, quantization demonstrated a significant reduction in model size by almost fourfold and a 3.3-fold decrease in RAM usage. The marginal average F1 score decrease was limited to only 1%. On the other hand, knowledge distillation resulted in a reduction of model size by 3.7 times, RAM usage by 1.9 times, and inference time by 1.7 times, accompanied by an accuracy decrease of 8%. The combination of both techniques yielded substantial outcomes, including a remarkable model size reduction by a factor of 14.6, approximately 4.3 times lower RAM usage, and a notable 2.4-fold improvement in inference time. Our compact model is the first end-to-end speech-based toxicity detection model based on a lightweight transformer model suitable for deployment in physical spaces. The results show its feasibility for toxicity detection on edge devices in real-world environments.