🗣 Edge#133: Self-Supervised Learning for Speech
Despite challenges, SSL proved to be an effective technique for ASR models
In this issue:
we discuss Self-Supervised Learning for Speech;
we explore AVID, an SSL model for audio-visual tasks;
we overview s3prl, an open-source framework for SSL speech models.
💡 ML Concept of the Day: Self-Supervised Learning for Speech
Most of the biggest breakthroughs in self-supervised learning (SSL) have come from the natural language processing (NLP) space, but several other deep learning domains are quickly catching up. Among those, speech analysis is one of the areas that regularly pushes the boundaries of SSL research. Traditionally, automated speech recognition (ASR) systems relied on the largest possible training datasets for different tasks. SSL has become a viable alternative to generalized knowledge using loosely labeled datasets.