2010-07-20

Translations into Noise

I was around when Jesus Christ had his moment of doubt and pain
-The Rolling Stones


Moments of doubt around me Jesus was a pain
-Google Translate, two iterations between English and Japanese


Around me in a moment of doubt and pain, Jesus was
-Google Translate, fourth iteration


Indeed, the pain, I was around the moment of Christ
-Google Translate, sixth iteration (equilibrium)


Translation Party provides endless fun, indeed.  We are familiar with the convenience of Google Translate.  We laugh at the slight mistakes that it makes.  An entire current of humor can be extracted from the overlapping missteps of a translation machine.  Yet there remain untapped regions for exploration here.  


A few months ago, Google upgraded its text-to-voice feature to 34 languages.  Using the open source speech synthesizer espeak, we can now hear the monotoned synthesizer babble out our oft ridiculous translations in Swahili, Welsh, or Icelandic.  The espeak synthesizer works by compressing languages into spectral data, which allows for many languages to be stored and accessed quickly.  The lack of smoothness in the speech is made up for in the clarity of the signal.


It is this lack of smoothness, the choppiness of the translator's voice, that I find most interesting--the lack of transitions between sounds.  (There must be a term in acoustic phonetics for this) Testing the new translator,  a Thai friend of mine updated his Facebook status in Thai, which I copied into Translate and got out a bunch of nonsense.  But then I translated that nonsense into the first language on the list, Afrikaans, and recorded the sound.  Listen:












then I slowed it down:













slower still:













and slowest:
















When it's completely slowed down, you are hearing are the pure resonances in the machine's speech, also known as the formant.  A formant is a spectral peak, easily detected, for example, when you make different vowel sounds.  We identify these sounds by their distinct signatures in the sonic spectrum.  






From the Wikipedia entry on 'formant', we might find our way back to architecture: "Most of these formants are produced by tube and chamber resonance, but a few whistle tones derive from periodic collapse of Venturi effect low-pressure zones." Referring to tubes and chambers in our oral passages, could we also have cave- and tube-like spaces in buildings which generate a variety of spatial formants?  Perhaps buildings could not only speak in the Robert Venturi sense but also in the Giovannia Battista Venturi sense of air squishing and flushing through its hollows.  Buildings could have conversations.


In 1969, Alvin Lucier was pondering a room's static sonic signature.  Famously, he recorded the sound of his voice and played it back over and over until the content was muffled and the resonant tone (formant) of the room was about all that was left.  If you listen to the recording in the above link to UbuWeb, you might fast-forward to the end which will just sound like noise.  I recommend listening to the 15 minutes straight-through.  Listen to the process of the room's resonance overtaking his stuttering voice.  Once it is understood that the noisy recording at the end is actually the sonic signature of Alvin Lucier's room, it ceases to be noise.  


The translation which seemed lost becomes the signal itself.