blog beats life
Google Voice
OS X Speech Synthesis
Ableton Live

“I propose this as a standard test for end-to-end speech recognition to speech synthesis: The 'G' Thang Benchmark.”

Not quite there.   

“One two three and to the fo’ / Snoop Doggy Dogg and Dr. Dre is at the do’ / Ready to make an entrance so back on up / ‘Cause you know we ‘bout to rip shit up” = Snoop & Dr. Dre

“Like cos I was like bath and stuff like that.” = Google Voice’s Remix

I love Google Voice. Admittedly, I primarily use it (and GrandCentral before that) as a modern-day door opener (more on that below).

I recently got the urge to call up my GV number and leave a message to see how well it would translate the ‘a cappella’ version of Nuthin’ but a ‘G’Thang.

A proposal: The ‘G’Thang Benchmark

As I see it, there must come a time when not only will this be nearly translatable, but I will receive voice-analysis meta-data along with the transcription that will indicate that indeed, Snoop and Dr. Dre just left a message for me.

I could then take the high-fidelity transcription and use it to generate a high-fidelity vocal rendering, using a pair of $.99-cent OS X voice-synthesis packs I purchased on iTunes that provide me with accurate vocal emulations of the two famous MCs.

Finally I would drop the generated vocals over the original instrumental track and compare to the original track from 1993. I propose this as a standard test for end-to-end speech recognition to speech synthesis: The ‘G’Thang Benchmark.

Alas, the days of Snoop reading us our morning newsfeeds is far away; for the most part the translation is a far cry from the original, although there are some funny interpretations:

"And before me dig out a bitch / 
I have ta' find a contraceptive"		

"And before I get a chance. /
I have to find the cats or step. If"


"And who gives a fuck about hoes / 
So just chill 'til the next episode"

"But anyway, I got off the flow /
of just change but I think that's so"



"That's hooked on that sucks blood /
on the ball ball because of the brick at the house, /
but it was destroyed it. /
And they they call the phone. /
Please put the sprint. So it looks like a home, /
which is also just to look at it /
would but I don't like that /
that and it benefits like that. /
Like cos i was like bath and stuff like that."


Footnote: You see, I still have an out-of-state cell-phone number — the same Massachusetts digits that birthed my first cell-phone experience on a Motorola StarTAC, somewhere around 1999.

My San Francisco loft has a front-door callbox that will ring the home phone to let visitors be buzzed in. However, the callbox would only take local numbers and I haven’t had a land-line in almost a decade!

The solution presented itself once off the wait-list for GrandCentral: an hour of weeding-out Bay Area (415) numbers containing a 1 or 0, then cross-checking potential candidates with PhoneSpell.org, yielded a memorable phrase + local number.

Now when I am too lazy to take out my keys, I buzz myself in, involving an unfathomable chain of modern technologies and physical distances.