Biopartnering Seminar - Developments and Advancements in Speech Recognition Technology

Symbiotec - Dec 2nd
Tuesday, December 2, 2014 - 6:30pm

Speech recognition is defined as a type of pattern recognition that translates a speech signal input into relevant mediums, such as transcribed text. This technology is composed of four blocks, each of which are independently configured and trained: the acoustic model, the language model, the decoder, and the pronunciation dictionary. Although they are trained separately, their performances are generally measured against a common standard called the Word Error Rate, a test that quantifies the system’s ability to accurately recognize basic spoken language units called phonemes.

The inputs are processed using algorithms and machine learning techniques, such as Hidden Markov Models (HMM) and Deep Neural Networks (DNN). HMMs are graphical statistical models that measure the probability of the next phone based on previous phones. They are currently used in conjunction with DNNs, which are mathematical neuron-based models, to approximate outputs based on a large number of inputs.

Despite recent advancements in both computing power and labelled data availability, the state of the art faces several limitations impeding the widespread application of speech recognition, such as issues with extracting semantics and emotions within the language model. Advancements in natural language processing are expected to overcome these limitations. Furthermore, Improvements can also be expected from modifications in machine learning methods to replace older statistical models with newer DNNs.

This talk will explore the evolution of speech recognition technology, including a breakdown of the recognition process and relevant technological advancements in the field, which will permit its widespread adoption.

Team: Symbiotec

Team Members: Jonathan Barr, James Boam, Emilie Colton, Jeffrey Kwesiga, Rachel Suliono

Location: Instructional Centre, Room 120

All welcome! Light refreshments will be served at 6pm.