Talk To Me

Ever since its first baby words in 1984 the Mac has been a garrulous creature. Vintage users will remember with varying degrees of hilarity or annoyance the Talking Moose b. 1986 who would pop up randomly and add a pithy comment to lighten up a dreary work session. And he's still available. But be warned - he's an absolute time waster. If this piece has some weird grammar - blame the Moose! In fact I've just had to turn him off.
In 1993 I invested in a Centris 660AV which also talked back. More importantly, I could talk to IT. In a rudimentary sort of way. " Open Claris Works." "Quit Word." And sometimes it would.
Since then we've had to wait a long time for good voice recognition on the Mac. IBM's ViaVoice around 2003 did a lot of heavy lifting for me but strangely stopped working across a number of machines as Tiger progressed through its maturations. iListen which used a different technology, showed a lot of promise but didn't particularly like my voice. Maybe it's because, even after 50-odd years in Oz, I still can't settle my vowels between the local nasal twang and the flattened sounds of my native Yorkshire. Ee 'oop, lad! But now MacSpeech have licensed the Dragon speech recognition engine which for years has delivered a rare and galling superiority to our Windows brethren. So now we have Dictate for the Mac which by all accounts is the best yet by far. But more of this in a coming edition.
The Mac system voices didn't show much development beyond the novelty factor until recently. Although Leopard's Alex sounds pretty good and has developed somewhat of a cult following on YouTube, there are some other impressive voices out there. Recently I had a listen to offerings from AssistiveWare, who distribute the Infovox iVox system from Acapela, and Cepstral. Infovox have a range of male and female voices with fairly neutral English and American accents, and also cover most European languages. Cepstral offer fewer languages but have more variations on a particular language. Demos are freely available. But it's not only the sound of the voices that has improved in recent times. Greater processing capacity has allowed deeper sampling which produces more natural speech patterns and intonation. The voices are still a bit wooden and clunky at times but you'll also get fluent passages where you almost forget that it's a Machine.
As a long-time teacher of Languages other than English (LOTE) I was particularly keen to try the German, French, Italian and Spanish voices. My first impressions were all positive and I would recommend to my LOTE colleagues and particularly their students to grab the demos and try for themselves. Lots of possibilities here.
The voices download into the Speech pane of System Preferences where you can listen to short samples as you vary the speed to suit. The UK voices are particularly refreshing after decades of computerised American voices. But when Lucy the Pom read this piece to me did I detect a faint note of disapproval in her tone? Her US colleague Allison seemed to sound more enthusiastic about what I've written.
But there's even more in the wind. I recall sometime in the early 90s a company which claimed it could produce a unique font from samples of your handwriting. Acapela now offer something similar by building a custom voice based on a real person's speech patterns. Now that's got to be something worth keeping an eye, or rather an ear on. It would sure beat having to do, say, John Lennon, by ransom-noting sound snips of his voice into an original utterance - which he probably never uttered anyway. If you really wanted to.
In another development the Acapela/Infovox vocal stable have introduced a phonetic correction facility to deal with mispronunciations. This looks quite promising and I'll report on it next time as well as looking at the various voices in more depth.