top bar
QuickTopic free message boards logo
Skip to Messages


Comments and discussion on TBTF for 1999-10-05: Offlist

Barry Jaspan
12:22 PM ET (US)
I have a comment on the article regarding the neural network voice recognition that is possibly better then human voice recognition. In particular, the "better then human voice recognition" part. I'm not really agreeing or disagreeing with the article, just making what I considered to be a fascinating and not widely recognized point.

I use Dragon NaturallySpeaking voice recognition software (in fact I'm using it to dictate this message). This product is designed to perform a specific task: transcribe spoken words into printed text. As such, it is competing against two existing "technologies":

1. Writing/typing by the actual author of the document directly to paper/typewriter/computer, and

2. Dictation by the actual author to a human within transcribes the document to paper/typewriter/computer.

The amazing fact is that Dragon NaturallySpeaking (and at least some of its competitors) is ALREADY superior at transcribing full speed, large vocabulary general purpose English prose than all but the most highly trained humans using either of the two alternative technologies listed above. Voice recognition software can transcribe text faster than just about anyone can type, although once you consider the time required to correct voice recognition errors it would merely require a "fairly fast typist" to do better (but a "fairly fast typist" is a pretty highly trained person). And, obviously, it requires a court reporter or highly skilled secretary to take dictation from another person in anything approaching real-time. Undoubtedly, the current state-of-the-art in commercial voice recognition software (which costs $170 and requires perhaps a $1000 computer) could beat just about any reasonably intelligent high school student at both raw speed and accuracy in recognizing and transcribing a single human voice.

Of course, the human would have approximately the same capability with a wide variety of voices, be able to adapt to new speech patterns more quickly, work under noisier conditions, be able to guess the spelling of words never heard before, etc. So humans are still in general better than software. But, as in chess, the computer's ability at a task once thought to be entirely in the realm of biological intellect has been demonstrated as superior at least within certain well-defined (but nonetheless useful and practical) constraints.
Keith Dawson
10:41 PM ET (US)
Discovered a URL error in the TBTF email edition (just after it was sent, natch). If you follow one of the links from the email looking for the Take It Offline discussion you will end up here. The link cited directly below is correct for the TIO discussion.

Hate it when that happens.
Keith Dawson
01:28 PM ET (US)
Use this forum to comment on or discuss TBTF for 1999-10-05: Offlist. I've set up other forums for three of the articles in this issue; if your contribution is about one of these please follow the link.

- The Take It Offline service and its impact on the dynamics of email lists and discussions

- Ken Thompson's paper "Reflections on Trusting Trust"

- Easter eggs built into search-engine rules

Print | RSS Views: 863 (Unique: 456 ) / Subscribers: 0 | What's this?