Alex: An Unsung Hero of Leopard

If you haven't already, try out the new synthesized voice in Leopard called Alex. The easiest way to do so is to use the 'say' command line tool with a string you'd like him to speak.

Download a sample (72k) of this text:
"I can say some fairly complex things. iMac. iPhone. apple.com. Theocacao."

It's not just the quality of the pronunciation that's impressive, but the low amount of distortion and the ability to use real-world phrases correctly. You can also adjust the rate of speech in System Preferences → Speech → Text to Speech.

Speech Rate


You can output the result of the 'say' command using this syntax:
say -o file.aiff "Hello again."

Let this put to rest the question of how to pronounce Theocacao. I don't know, but Alex apparently does.
Design Element
Alex: An Unsung Hero of Leopard
Posted Dec 18, 2007 — 15 comments below




 

Jesper — Dec 18, 07 5223

You see, this is why I named my site "Waffle".

Baiste — Dec 18, 07 5224

And he actually breathes. This may not seem like a big deal (it's fairly easy to implement), but it caught my attention right away. He just feels more human, in a subtle way.

Steven Fisher — Dec 18, 07 5225

This sounds interesting when spoken: "the low amount of distortion"

Jose Vazquez — Dec 18, 07 5226

Thanks for bringing it up. I just spent 20 minutes playing with my 3 year old daughter. She is a spanish speaker, but she knows a few things in english (colors and what not). She had a blast hearing the computer talking. I actually had a mini conversation with her via the laptop.

Jane — Dec 18, 07 5227

From my experience, Alex's lovely for Voiceover until you jack up the speaking rate to something fast...and by fast I mean a bit faster than average speech, cause it's quite annoying to have to read a whole webpage or something slowly. On the flip side, if I'm reading a book for leisure and feel like using the voice to read it to me at a slower rate, it almost feels like a real person is talking.

So, it's fantastic for what it is (free voice that comes with the OS, natural sounding and all)...but it's not that fantastic compared to some of the voices available out there.

Justin Williams — Dec 19, 07 5228

Alex rocks. After I'm done writing my column, I highlight all the text and let OS X speak the text. It sound real enough where I can easily notice any bad wording or missed spelling errors.

Matt J — Dec 19, 07 5230

Alex is pretty good, and pronounces my last name better than some humans, but really messes up saying 'Crumpet'.

Blain — Dec 19, 07 5231

And he actually breathes.
I was about to note the same thing. I tried:
say "Peter piper picked a peck of pickled peppers, and ate them quickly. It was a big meal, with pickled peppers."
I had to play it several times to make sure I wasn't hearing thing, and actually heard him take in a breath between sentences.

Dmitry Chestnykh — Dec 19, 07 5233

English test:

say "Am I Australian: can't; or British: either."

Kevin — Dec 21, 07 5251

But while the voices have improved at some point (prior even than tiger) they dumbed down the parser: With the old Apple speech parser this sentence would be spoken correctly: "Did you read the book I read?" pronouncing the first rEEd and the second red; Now there parser has no sense of tense. There are other examples of encroaching retardation as well but I just got informed that the coffee shop I'm in is closing so, well, I guess I better just end here.

Chuck — Dec 21, 07 5260

@Kevin:
I'm intrigued that it could correctly parse that sentence before, because there are almost identical sentences that I'm not sure I could correctly read without more context. For instance, "Did you read the same magazine I read?" This sentence makes perfect sense with the second "read" as either present (meaning I habitually read the magazine) or past tense (I once perused this magazine). I wonder how smart the old parser was.

Kevin — Dec 22, 07 5270

But while the voices have improved at some point (prior even than tiger) they dumbed down the parser: With the old Apple speech parser this sentence would be spoken correctly: "Did you read the book I read?" pronouncing the first rEEd and the second red; Now there parser has no sense of tense. There are other examples of encroaching retardation as well but I just got informed that the coffee shop I'm in is closing so, well, I guess I better just end here.

Kevin — Dec 22, 07 5271

@Chuck,

Point taken (since I can't prove this exact sentence); it is an example of me trying to be quick and thus changing a key variable. This is the test phrase I used to use: "I just read a new book; did you read it?" My point was that the parser used to recognize tense with "read" and it no longer does. In fact, Read is, now, always read as rEEd. I don't know when the change occurred but I can say that even in OS 8, the parser would correctly guess tense on the word Read.

Note also: "Did you read the book I read" is faulty grammar if read is pronounced rEEd (because of ambiguous tense, and/or the nominative noun): it could be "did you read the book I am reading" (referencing a specific book in the present tense) or "do you read the book I read" were the book is a nominative noun. The only proper translation of "did you read the book I read" is in the past tense and thus pronounced red.

PS -- sorry about the double comment; I refreshed the page.

Ricket — Jan 30, 08 5414

Kevin, your grammatical analysis is faulty itself. A /reed/ reading that would make grammatical sense: "Did you read the book that I (always) read (on Sunday afternoons)?"

Further, one could argue that your "corrected" example is likewise ungrammatical: "Did you read the book I am reading?". Shouldn't that be "Did you read the book that I am reading?" ?

I'm no prescriptivist, but I think that those who are need to tread carefully.

Naseer Khan — Mar 09, 08 5604

Hmm... I think the main point of any speech synthesis is to make a computer sound like a human -- of contemporary means. Proper grammar aside, the idea is that there are different forms of pronouncing the same word (hey, who made this language?)... and any smart computer should be trained to separate colloquial distinction. All sides agree that this is a mess but isn't that what makes it a challenge...!!!

Alex is a great improvement, but it can't be said to be the end of the road, especially the way Apple pumped it up in the unveiling. Besides it can't even pronounce my name correctly.

--Naseer (try "Nuhcier")




 

Comments Temporarily Disabled

I had to temporarily disable comments due to spam. I'll re-enable them soon.





Copyright © Scott Stevenson 2004-2015