A Review of Deep Learning–Based Sarcasm Detection in Monolingual Speech Using Sentiment Analysis

Abstract

Sarcasm detection in spoken language has emerged as a challenging problem in affective computing and speech analytics due to its reliance on implicit meaning, contextual cues, and paralinguistic features. In monolingual speech settings, sarcasm is often conveyed through subtle variations in prosody, tone, pitch, and rhythm rather than explicit lexical markers, making traditional rule-based or shallow machine learning approaches insufficient. Recent advances in deep learning have significantly improved sarcasm detection by enabling automatic feature learning from raw audio signals and speech-derived representations. This review critically examines deep learning–based approaches for sarcasm detection in monolingual speech with a specific focus on sentiment analysis–driven frameworks. It discusses commonly used architectures such as convolutional neural networks, recurrent neural networks, long short-term memory networks, and transformer-based models, highlighting how they integrate acoustic, prosodic, and sentiment-oriented features. The paper also reviews widely used speech corpora, evaluation metrics, and preprocessing techniques relevant to monolingual sarcastic speech analysis. Furthermore, key challenges—including data sparsity, speaker dependency, contextual ambiguity, and class imbalance—are identified, along with emerging trends such as multimodal fusion and self-supervised learning. This review provides a structured synthesis of current research and outlines future directions for building more robust and context-aware sarcasm detection systems in monolingual spoken discourse.