Speech Technology Magazine

Published Fall 2018 - Letters to the Editor

The Price of Imperfection
In Speech Technology’s Winter 2018 issue, several articles discussed billions in potential and lamented the lack of perfection restraining that potential. This was not new, but the frustration appears to be reaching new highs. Worse, nothing is changing, and the potential remains out of reach.

In 2012, “From Speech Signal to Phonological Features - A Long Way (60 years and counting)” (from The Journal of the Acoustical Society of America) was presented at an academic conference. This title succinctly states 60 years of imperfection, and we are 4 years from a 70-years-and-counting presentation.

Business R&D makes advances, but specifics are proprietary and have limited dissemination. We do know when research bears fruit - a business is born or bought, and another piece of the puzzle becomes available to a small few.

As a business, Waveform Communication has announced several innovations in speech technology. Just as important, the perfection achieved by the Waveform Model of Vowels (WMV) has been presented at academic conferences. Although this is a unique approach, this has been done to push through over 60 years of noisy failure.

This is not an indictment on the fantastic and difficult work that has been done. The fact no one completed the puzzle before is partially a reflection of the research and business relationship described above. Developing a working model of cognition is difficult enough without missing or incomplete pieces.

Trivial criticisms of the writing have been made, but nothing regarding a flaw in the logic of the WMV.

Working computer programs utilizing the WMV are an extraordinary defense of the logic. How many points are needed to win a debate, and when is the debate of competing models going to happen?

This is not hubris. The WMV has achieved perfection for years and presented it publicly. There are a few things for the speech community to consider. Do others expect to find an equally effective method as the WMV that also explains human perception, production, and errors? Are businesses willing to continue to misidentify vowels across talkers? As a reminder, other methods are at 66 years and counting.

We know the price of imperfection. Embracing perfection seems like a logical alternative.

Michael Stokes
President and CEO
Waveform Communication, LLC