Acoustic and Speaker Modeling Using Deep Learning

There is a rapidly growing need for voice powered human-machine interaction modalities for varieties of devices. Despite enormous investment in research and development in this area by a number of companies, significant limitations remain which prevent the ubiquitous proliferation of speech recognition. These limitations include poor performance in the presence of noise, inability to handle variability in accents, and not reliably recognizing the speaker. Fluent.ai is investigating novel neural network architectures for solving aforementioned issues with a primary target market of personal smart devices such as wearables, smart-toys and smart-home devices. Two key aspects of a successful voice user interface solution for such devices include high recognition accuracy even in the presence of noise and ability to adapt to speakers with difference speaking styles and accents. It is crucial for Fluent.ai’s business that these challenges are addressed effectively. TO BE CONT’D

Faculty Supervisor:

Patrick Cardinal

Student:

Mohammed Senoussaoui

Partner:

Fluent.AI Inc

Discipline:

Visual arts

Sector:

Information and communications technologies

University:

Program: