AES Show 2024 NY

The aim of this workshop is to provide participants with hands-on expertise in utilizing machine learning for audio quality modeling.

This workshop is a follow-up from the initial edition presented at the 156th AES Convention. The workshop expands on its first edition by including the participation of new panelists working at cutting-edge technology with multimedia industry leaders. Particularly, very recent advancements on deep learning for auditory perception will be presented, as well as a new open audio quality dataset for informing auditory models in audio quality metrics.

Machine learning techniques have been instrumental in understanding audio quality perception, revealing hidden relationships in subject response data to enhance device and algorithm development. Moreover, accurate quality models can be used in AI systems to predict audio quality -- a critical part of customer experience -- in situations where using human subjects is costly (e.g., in day-to-day product development) or impractical such as in audio transmission network monitoring and informing deep learning audio algorithms.

The complex nature of quality perception requires users and developers of these models to possess specific domain knowledge that extends beyond the general machine learning set of skills. This expertise includes experimental design in the domain of subjective audio quality assessment, data collection, data augmentation and filtering, in addition to model design and cross-validation.

The workshop aims to shed light on historical approaches to addressing these challenges by marrying aspects of machine learning and auditory perception knowledge. Furthermore, it will provide insights into state-of-the-art techniques and current related research topics in data-driven quality perception modelling.

The main goal of the workshop is to offer its attendees —- whether users or developers —- the necessary tools to assess the suitability of existing ML-based quality models for their use case.