Biologically-inspired auditory artificial intelligence for speech recognition in multi-talker environments

Understanding speech in the presence of distracting talkers is a difficult computational problem known as the cocktail party problem. Motivated by auditory processing in the human brain, this thesis developed a neural network to isolate the speech of a single talker given binaural input containing a target talker and multiple distractors. In this research the network is called a Binaural Speaker Isolation FFTNet or BSINet for short. To compare the performance of BSINet to human participant performance on recognizing the target talker's speech with a varying number of distractors, a "cocktail party" dataset was designed and made available online. This dataset also enables the comparison of network performance to human participant performance. Using the Word-Error-Rate metric for evaluation, this research finds that BSINet performs comparably to the human participants. Thus BSINet provides significant advancement for solving the challenging cocktail party problem.The research was funded by an NSERC Canada Discovery Grant, a Government of Alberta Centre for Autonomous Systems in Strengthening Future Communities grant, a MITACS Globalink Award, a NSERC CGS-M Award, and a AITF Graduate Student Scholarship