Recently, neural networks have shown impressive progress across diverse fields, with speech processing being no exception. However, recent breakthroughs in this area require extensive offline training using large datasets and tremendous computing resources. Unfortunately, these models struggle to retain their previously acquired knowledge when learning new tasks continually. In this paper, we investigate the problem of learning sequence-to-sequence models for spoken language understanding in a class-incremental learning (CIL) setting and we propose COCONUT, a CIL method that relies on the combination of experience replay and contrastive learning. Through a modified version of the standard supervised contrastive loss, COCONUT preserves the learned representations by pulling closer samples from the same class and pushing away the others. Moreover, we leverage a multimodal contrastive loss that helps the model learn more discriminative representations of the new data by aligning audio and text features. We also investigate different contrastive designs to combine the strengths of the contrastive loss with teacher-student architectures used for distillation. Experiments on two established SLU datasets reveal the effectiveness of our proposed approach and significant improvements over the baselines. We also show that COCONUT can be combined with methods that operate on the decoder side of the model, resulting in further metrics improvements.

This paper describes a multichannel acoustic data collection recorded under the European DICIT project, during the Wizard of Oz (WOZ) experiments carried out at FAU and FBK-irst laboratories. The scenario is a distant-talking interface for interactive control of a TV. The experiments involve the acquisition of multichannel data for signal processing front-end and were carried out due to the need to collect a database for testing acoustic pre-processing algorithms. In this way, realistic scenarios can be simulated at a preliminary stage, instead of real-time implementations, allowing for repeatable experiments. To match the project requirements, the WOZ experiments were recorded in three languages: English, German and Italian. Besides the user inputs, the database also contains non-speech related acoustic events, room impulse response measurements and video data, the latter used to compute 3D labels. Sessions were manually transcribed and segmented at word level, introducing also specific labels for acoustic events.