I don't understand why this is a problem. Apple provided great example voices with very good bells and whistles in their Speech Manager API. And still, voice developers skip over some fundamental functionality. My pet peeve is punctuation marks.
When a text-to-speech voice comes to a punctuation mark, you want it to act accordingly like a rational human being. If a sentence ends in a period, the voice speaks in a 'straight' tone. If a sentence ends in an exclamation mark, you want the voice to speak more emphatically. If a sentence ends in a question mark, you want the voice to rise at the end in a questioning tone. All of Apple's voices do this.
Cepstral voices treat exclamation marks and question marks as periods. It sounds very odd and dysfunctional. This is why I avoid their voices.
Acapela voices, which are used in Infovox, are just as nice sounding as Cepstrals with the added feature that at least they (usually) comprehend question marks and speak accordingly. (Results will vary, so test voices before using them). But again, sadly, they have no comprehension of exclamation marks, which of course sounds again odd. Bleh. But at least they're a step up from Cepstral.
Then there is the noise artifact problem. Apple's recent Alex voice still has them. You hear odd little noises that have nothing to do with speech. The Bruce voice, in one version of Mac OS X, used to make occasional farting noises! Vicki remains the queen of Apple voices for lack of artifacts. But sadly the Leopard and Snow Leopard speech engine ruins her cadence to make her sound manic, despite the fact that nothing whatever changed in her actual speech code.
Cepstral and Acapela voices can have similar noise artifact problems. Be sure to check each voice carefully before buying it or deploying it. One voice I with which I am most pleased is the recent Peter British voice from Acapela. I find it easy to understand with very few noise artifacts or speech inflection oddities.
It's strange after all these years of voice technology, [The original 1984 Mac had text-to-speech built in], that it's still 'buyer (or user) beware'. No doubt human error inflicts ever inflicts itself into the process.
Infovox iVox
Natural voices in language of choice for use in any app.
Version: 2.0
Waiting for Cepstral update ... (german voice) - jopi1
As there is no snow leopard update for Cepstral by now, I had to change from "Matthias" to "Klaus". I really prefer the more neutral Cesptral voice, long texts with cheerful "Klaus" are painful.The Pronunciation Editor with InfoVox is a great tool, I never got into this with Cepstral.
I could even stop InfoVox reading wikipedia quotations like this: "Anführungszeichen oben .... Anführungszeichen unten" by changing the phonetic pronunciation to "#_" This might work with “asterisk asterisk asterisk asterisk asterisk asterisk asterisk asterisk asterisk” as well.
Hopefully the 30 days demo will be okay for Cepstral to fix Matthias:)
Reply to This
Tuesday, September 29 2009 @ 11:53 PM PDT