19AES-PD14: Deep Learning and AI for Audio Applications - Engineering Best Practices for Data

Presenter:
Gabriele Bunkheila, MathWorks - Madrid, Spain

Audio, speech, and acoustics are increasingly recognized as the second largest application area for deep learning after computer vision. Deep learning and AI are defining a new era in product development as they need vast amounts of task-specific labeled data to be successfully optimized for real-world applications. As deep learning is increasingly used alongside more traditional signal processing methods, the focus of audio DSP engineering is gradually expanding from algorithms to data.

In this session, we discuss the importance of signal processing and audio data engineering for the development of deep learning systems. Using practical examples based on MATLAB, we review best practices for audio data workflows in AI applications, including for signal labeling, data ingestion, data augmentation, feature extraction, and signal transformation.