0.3 How the speak n spell worked  (Page 4/6)

• Wordy Birdy
• Talking Typewriter
• Spelling Bee
• Franken-speller (my favorite)
• Robot

We were amazed how well he could put a voice to the various characters for the focus groups and management demonstrations. However, the professional speaker who could do all of the character voices was not LPC friendly so we could not use him for the final product. Figure 4 shows several of the product concepts. Unfortunately I lost the concept drawing for the Franken-speller.

Much to our surprise, we found that female voices also did not work at all with our synthesizer. We did, however, finally find a good LPC friendly voice in the Dallas area. But before we proceeded we did a quick market survey among children as to their preference of gender in synthetic voice. We found that a male voice would be acceptable to both boys and girls. This may have been our greatest technical hurdle of the program.

Later in the program, as we were qualifying the spoken words for the product we came across several issues as a result of the speech quality that I will assign to the decisions we made on sample rates. As I said earlier we chose to use an 8kHz sample rate which was fine for capturing the formants in a voiced sound. But it didn’t do so well with many unvoiced sounds that have the bulk of their information in the 2kHz to 6kHz frequency band. We could only capture frequencies less than 4kHz. In fact, as it worked out, we had a cut off at about 3.4kHz, but I’ll talk about that later, when I talk about the output speaker and power amplifier.

Specifically the unvoiced sounds f, s, t, p, h, sh, were sometimes misinterpreted as one of the others. Here are some examples:

• The word “four” was easily misunderstood for “hore” if not spoken in context. We needed to keep the number “four” but chose not to use any of its homonyms in the vocabulary list (e.g., for, fore).
• In a later product we realized that “p” and “t” could be misunderstood causing a specific problem with the word “ship”. We ended up canceling the product because of this issue.
• Other words like “ditch”, “sheet”, . . .

One of the games on the final product was created for this very concern – the “Say it” mode. We concluded that it would be wise to have a mode where the child could hear the word while looking at the spelling in hopes that when the word was spoken in the “Spell it” mode it could be easily correlated to one of the spelling words seen in the “Say it” mode. We later found out that this was liked by educators as it allowed the child to relate spoken words to written words.

While I’m on this subject, I will relay a quick story. After the Speak N Spell had been introduced, I was told that one of our Board Members said “I guess this is adequate speech quality for the initial product. But if any of the add on modules have this bad quality, we need to take the whole team, line them up in front of Neiman Marcus and shoot them.” As I didn’t actually hear this myself, it is just a good story to tell. And, yes, we did do follow on modules and products. The quality obviously improved as not one of us was shot in front of Neiman Marcus.