0.8 Taking the speak n spell into production  (Page 2/5)

We found that the professional speaker needed to be a male with a relatively low voice with little character. That is not “sing-songie” or a lot of vibrato. We called it an LPC voice.

With the male voice chosen (a DJ from the Dallas area), we began recording the words and editing them. Each word had to be edited to fix errors made by the computer while analyzing the sounds. Most of the errors were voicing errors, pops or ringing tones. Or at least we thought that was all. We first chose engineers and scientists to do the editing but quickly realized that the reason we were engineers and scientists was due to our lack of language skills. We ended up hiring linguists and language majors to do the editing.

Even with the proper professional speaker and editor, we had problems. As it turns out, the more one hears the word the better it sounds. We, as should always be done, formed a committee to resolve the familiarity issue. We had about six people on the committee with the editor. Our job was to listen to the word, without knowing what the word was, and scoring it. Each would write down the word they heard and whether it was acceptable or not. If it was not acceptable, then a description of the issue was added.

As an example of the problem, in one of these QC sessions, all six of us wrote down the word we heard and all agreed that the word was acceptably spoken. When we compared notes, we found that each of us had written down a different word. Unfortunately, I can’t remember the word, but the experience added a step to the QC procedure. It also was a preview of the difficulties we would have on the production line with production people who were not professional linguists.

I’ve said all of this to get us to the issues on the production line with speech quality. The issues were

• The noise on the production line
• The qualifications of the individuals doing the speech quality test
• The tester's familiarity with the sounds after listening to them many times

Noise on the production line

We couldn’t eliminate the noise on the production line. We considered moving the testing to another location where it would be quiet. But, that would be very difficult logistically and was abandoned. Although, we had many meetings in a quiet conference room arguing over speech quality and volume (yes, I eliminated a volume control as it saved a dime or so).

We finally solved the problem by building Styrofoam “dog houses” for the testers. These dog houses had three walls and a ceiling. They were about a meter on a side and about a meter and a half tall. The tester sat on the production line inside the dog house facing its back wall. They spent eight hours a day seeing nothing but the inside of a white enclosure. It was a bit quieter but far from the optimal environment for a tester. Many comments and suggestions were written on the inside of the dog houses. Few of the suggestions were either possible or implementable. I don’t believe any psychological problems were caused.