The ITS Audio Quality Research Program has been pursuing a quest for accurate real-time measurements of speech quality and intelligibility for decades. In FY 2020, Principal Investigator Stephen Voran was honored with the U.S. Department of Commerce Gold Medal Award for leading the creation of WAWEnets, a breakthrough in objective measurement of speech quality and intelligibility. The expertise of Andrew Catellier, formerly with ITS, was integral to implementing a solution that used neural networks, a machine learning technique, to achieve this breakthrough. Andrew and Steve trained neural networks using a large speech database to develop software that allows for quick and accurate assessments of speech quality and intelligibility. This paves the way for optimizing user experience and conserving radio spectrum.
Wideband Audio Waveform Evaluation Networks, or WAWEnets, do not need the original speech signal and thus can measure telecommunications systems while they are in use. They remove the constraint of only being able to measure speech quality in a lab or when equipment is out of service, enabling formerly impossible testing—that is, when devices attempt to remove noise (or otherwise enhance speech signals) and no original speech-only signal exists, so testing of the transmitted result is the only viable option.
The team leveraged recent advances in deep neural networks as well as a significant trove of data that ITS generated specifically for this purpose over the course of many months to achieve this breakthrough. The work takes advantage of expertise that ITS has accumulated over decades of working with public safety to improve speech intelligibility. WAWEnets facilitate moving accurate speech testing out of the laboratory environment, advancing innovation to ensure spectrum is available for federal and commercial services. It enables deployment of real-time, in-service, and accurate speech quality or intelligibility monitoring anywhere in a telecom network.
The WAWEnets software is available on NTIA’s Github page for other researchers to build upon. Andrew presented a more detailed description of WAWEnets at ICASSP 2020, the 2020 IEEE International Conference on Acoustics, Speech and Signal Processing: Andrew A. Catellier and Stephen D. Voran, “WAWEnets: A No-Reference Convolutional Waveform-Based Approach to Estimating Narrowband and Wideband Speech Quality.”