3/22/2016 Building on Ray's MIT Workby John Makhoul, BBN Colleague
For his Master's thesis, Ray built a computer-controlled speech synthesizer. In those days, speech synthesis was performed by exciting a filter bank with either a sequence of pulses for voiced sounds or with white noise for unvoiced sounds. The filter bank shapes the spectrum and produces the various sounds. So, the heart of the synthesizer was the filter bank. This was an analog filter bank in which the frequency of each filter was computer-controlled. Before I did my PhD thesis, around 1967, I was given a job to use Ray's design to build a speech synthesizer that used operational amplifiers (which had just come out) instead of an analog filter bank. So, this was a completely digital implementation of Ray's synthesizer (I did not actually build a filter bank, but rather a single digital filter that was computer controlled.) The big advantage of the new design was size. Ray's filter bank occupied a whole rack of equipment, while the synthesizer I built fit in a single slot on a rack. Ray always joked with me about how, in a relatively short time, technology can outdate what you did just a short while ago. The silver lining of this story is that, because of my work on the synthesizer, I was very familiar with Ray's filter bank. So, when I was doing my PhD thesis in speech recognition, I wanted to do close to real-time recognition of single words, but there was no way I could do that digitally; it would have taken way too long. Plus I was limited by memory: the single-user computer we had was a DEC PDP-9 that had only 64kB of storage only! So, my solution for doing spectral analysis of the speech was to use Ray's filter bank! Because it was analog, the spectral analysis was immediate! So, without Ray's filter bank, I could not have done that particular speech recognition thesis. For that, I was very grateful to Ray. Comments are closed.
|
Memories of Ray TomlinsonSign up to receive a notification when stories are addedArchivesCategories |