speech recognition | Fritz Bogott

iPod owners all eventually become subject to a pronoid form of apophenia, in which they become convinced that that the shuffle-play feature can read their minds.

Speech-recognition software users all eventually become subject to the paranoid version of this, in which they become convinced that the software is mocking them.

My father auditioned an early (1992-ish) cut of speech-recognition software intended for medical-transcription use. Whenever witnesses were present, the software interjected the word ‘testicle’ at least once every ten words. When witnesses were absent, this behavior completely disappeared. We eventually had to purge the word ‘testicle’ from the dictionary. Thereafter, if the dictation at hand required use of that particular word, the user was forced to spell it.

The software has improved immeasurably since then, but the curse has not lifted entirely. I have felt the clouds gathering for several days, and I quit today’s dictation session early when I spoke the words ‘pulled taut’ into the headset and the words ‘Pol Pot’ appeared on the monitor.

If mine were a more rational mind, I would doubtless ignore errors like these as statistical anomalies. But I am an engineer, and thus an implicit believer in black magic.

The good news: I wrote the first two scenes of the Pismo story today.

The bad news: I only wrote the first two scenes of the Pismo story today. There are fifty-six scenes in the outline and thirty days in June. That means I need to write an average of two scenes a day. Two scenes today: I haven’t banked any scenes for busier days to come.

The first scene was easy. I had already given it a lot of thought. The second was maddening. In my outline, I had written, “Exposition: describe the restaurant,” which wasn’t nearly enough meat for an entire scene. A sentence, maybe. I spent half an hour or so writing elaborate curses to myself until I invented enough business to build a scene around. I imagine that will happen a lot. Some of my outline notes are easy to write to; others not so much.

I also transcribed a few scenes from the longhand draft of the troll story using Dragon Naturally Speaking, which worked incredibly well. I used an earlier version of the software in the mid-1990′s to transcribe a book’s worth of interviews when tendinitis kept me from typing. Ten years seem to have brought lots of speed and accuracy improvements. Typing my longhand drafts is a chore. Dictating seems to take most of the sting out of it.

Fritz Bogott

Tag Archives: speech recognition

Apophenically Speaking

Day One