Using Machine Learning to Classify and Access an Audio Database

Voices.com, the largest online marketplace of voice talent, have identified Machine Learning as an enabler for future growth. In particular, incorporating Natural Language Processing (NLP) into structured queries and automatic classification of sample recordings. The first phase of this research will use machine learning to identify and train an NLP learnable parser. The second phase will be to automatically classify sound samples, which has been historically difficult due to low levels of accuracy. The approach will start with attributes for which accurate algorithms exist and new attributes will be added when usable levels of accuracy are achieved. Note that self-reported attributes are also not very accurate, so it is possible that automatic classification would beat user-reported attributes. The classification problem cannot be completely solved within the first year, but these classifications could speed up human processing or be used in the absence of human classification.

Faculty Supervisor:

Christopher Anand

Student:

Venu Kurella

Partner:

Voices.com

Discipline:

Computer science

Sector:

Information and communications technologies

University:

Program: