Friday, July 10, 2009

Google Voice Transcription

Here is one of my first Google Voice transcriptions.

i'm using the local voice compose this blog to describe my first experience was using google voice as the transcription machine so far it seems wonderful i have my office phone forwarded to my cellphone and then i have my crew over the phone number forwarded to my office phone that way i can both yet and have tran scott message is for various reasons the transcription it's very very fast i've been involved with speech processing at the end of this is 50 state university institute for signal and information processing for several years and seeing how complicated and complex the test of speech recognition is the groom translation pants your chin does not do 100% but they promised that it's going to get better actually belch and like everything else moving i suspect that would be true the transcription so for the same to post no limits the battery on your cellphone may go up before you were possibility since have a message automatically terminated calling well that's the first try let's see what happens

It appears like gobbledegook because there is no punctuation or sentence structure. That is OK for my purposes, since I only wanted a raw transcription that I will edit into a blog post. If I were listening to the recording instead of reading the transcription, there would be non-vocalized cues, such as durations of silence, to tell my listening processor the sentence structure. But after a couple of re-readings, I can glean the same information from the text.

My transcription of the same message is at the end of the article.

The first glaring omission is that the Google transcriber does not recognize my pronunciation of "Google" in three out of four instances. :) Someone who can put these two paragraphs into an automated comparison script will find a word error rate hovering around 40%.

I have suggested to the Google Voice Help system that they allow the user to edit the transcription, and then use the edited version and the recording as training input for future transcriptions. It could be tailored to the user, or added to the common speech vocabulary (especially, when the user uses a word not yet in the standard vocabulary).

My first take is that Google has another potential winner on their hands. Remember, this is a pre-release version. It will get better.

The human-transcribed message:

I'm using Google Voice to compose this blog to describe my first experience with using Google Voice as a transcription machine. So far, it seems wonderful! I have my office phone forwarded to my cellphone and then I have my Google phone number forwarded to my office phone. That way, I can both get and have transcribed messages from various people. The transcription is very, very fast. I've been involved with speech processing at the Mississippi State University Institute for Signal and Information Processing for several years and seen how complicated and complex the task of speech recognition is. The Google translation-transcription does not do 100%, but they promosed that it's going to get better as it goes, and, like everything else in Google, I expect that would be true. The transcription, so far, has seemed to pose no limits. The battery on your cellphone may go out before your ability to have a message automatically terminated. Well, that's the first try. Let's see what happens.