A bag-of-features framework for incremental learning of speech invariants in unsegmented audio streams

Abstract

We introduce a computational framework that allows a machine to bootstrapflexible autonomous learning of speech recognition skills. Technically,this framework shall en- able a robot to incrementally learn to recog-nize speech invariants from unsegmented au- dio streams and withno prior knowledge of phonetics. To achieve this, we import the bag-of-words/bag-of-featuresapproach from recent research in computer vision, and adapt it toincremental developmental speech pro- cessing. We evaluate an implementationof this framework on a complex speech database.