DV Info Net - Speach to Text Software?

I haven't tried the iPhone version but thought it was primarily a dictation app. (Also, we don't yet have iPhone coverage in the area where I live and work.) I have worked a bit with the Nuance/Dragon apps but found their usefulness was limited to dictation (where you can train them to your voice). Never had much luck with it for translating/transcibing meetings, interviews, etc.

But, if you've got Adobe's Premiere Pro CS4 or CS5 or Soundbooth CS4 or CS5, you've got a speech-to-text program in their "metadata speech analysis function" which can do a decent job with interview transcription.

The operative word is "may." I do a fair amount of work for lawyers. This sometimes involves video of depositions and sometimes mp3 or wav audio recordings of meetings and interviews. I've used the CS4/CS5 program when they've needed something so immediately --- say, when we are on the road or its at the end of the day and the secretaries have gone home and the lawyers need something to study that evening.

I import the audio or video file into CS5, put it on the timeline, click on the audio track, go to the source monitor window and click on the Metadata tab, and then click the "Analyze button." The last time I did this, it transcribed about 6½ hours of deposition into a text file in about an hour.

This could not replace a stenographer's transcription, but it was good enough to be usable for what the lawyer needed.

Adobe Speech Analysis does a very good job in distinguishing one deposition/interview voice from another for purposes of separating questions and answers. It gives you a script where it identifies the speaker (with a label/name) gives a text of what was said. The response/answer is identified as coming from another speaker. Although it does a good job of separating the text from different speakers, sometimes it identifies the same voice as a different speaker.

It has trouble with technical terms.

As you would expect, much depends on the clarity of the recording, the absence of ambient noise, and timbre of the speaker's voice.

Accents can throw it for a loop. I think of an example of this kind of thing from a Prairie Home Companion/Guy Noir episode. Guy is in North Carolina looking for a town called Boiling Springs. Minnesotan Guy (using a sort of New York accent) pronounces it "boyle-ng" but the North Carolina locals say "bye-lyn." It is hard enough for us to figure out what was meant when a Texan pronounces "cookie" with four syllables or a Minnesotan says, "you betcha' ... well, you get the picture.

So, "usable" and "accurate enough to immediately drop into a publishable document" can be different things. Proofreading and corrections are a must. The system can learn and can work with "reference scripts." Say, you have several interview sessions with the same speaker. You use the corrected first text conversion as a "reference script" for helping with accurancy in translating later interviews.

Actually, this software reminds me of what is was like with PC-based OCR from 15 and 20 years ago. Sometimes, the results were excellent. Sometimes, the scan worked well enough to be usable. Sometimes you were better off just typing the thing from scratch.

Hope this helps.