Audio Sentiment Analysis via Distillation and Cross-Modal Integration of Generated Multilingual Transcripts

2026-07-07

AI Summary

Existing audio foundation models struggle to jointly capture prosodic and semantic information in speech, limiting their performance in emotion recognition. To address this, this work proposes a cascaded cross-modal Transformer architecture that integrates raw audio with automatically generated multilingual ASR transcripts and their translations, enabling multi-level multimodal fusion. The knowledge from this multimodal teacher model is then transferred to a unimodal audio student model via knowledge distillation. This approach significantly improves accuracy in emotional polarity classification without increasing inference cost. Ablation studies confirm the effectiveness of both the automatically generated multilingual textual inputs and the distillation strategy, highlighting the novelty and practicality of the proposed framework for enhancing unimodal audio-based emotion analysis.

Abstract

Automatically recognizing the sentiment, positive or negative, from speech is a challenging task, requiring both the analysis of vocal inflections and the interpretation of uttered words. Recent solutions rely on audio foundation models to solve the task, but it remains unclear if such models can take all aspects into account. To this end, we propose a multimodal solution that integrates audio and text information via cross-modal transformers, where text transcripts are automatically generated via an automatic speech recognition (ASR) tool. Moreover, we create multiple text modalities by automatically translating the transcripts into multiple languages via machine translation tools. Audio and multilingual text features are combined via a cascaded architecture comprising cross-modal transformer blocks that integrate modalities one by one. We further distill knowledge from the multimodal model, called teacher, into a unimodal (audio only) model, called student. We conduct experiments on a large-scale dataset, demonstrating that the automatically generated textual information can bring significant performance boosts in multimodal sentiment polarity classification. Our ablation study confirms that both automatic transcripts and automatic translations are helpful. Moreover, we show that the audio-only model can be enhanced via distillation, boosting performance without any computational overhead during inference. To reproduce the reported results, we publicly release our code at https://github.com/andreidurdun/cross-modal-audio-sentiment.

Problem

• Audio Sentiment Analysis
• Multimodal Learning
• Automatic Speech Recognition
• Machine Translation
• Sentiment Polarity Classification

Innovation

• cross-modal integration
• knowledge distillation
• multilingual transcripts
• audio sentiment analysis
• cascaded transformer