Thursday, September 18, 2008

Producing a transcription with Dragon Naturally Speaking

This is my next experiment with Speech Recognition. This time I'm using Dragon Naturally Speaking version 10, which seems much more accurate than the engine built into Windows Vista. I didn't use Dragon as a dictation device, but rather to make a "transcription." Below is the raw text it produced, unedited. I recorded the audio with Audacity. That's also included. With multimedia where there is voice over of other content, and the voice track is produced after the visual track, I can now see producing a tolerable transcript as within reach. This is still far from perfect, but getting a tolerable result is entering the range of the do-able.

By the way, in training Dragon with my voice, there was an option for it to access my email and My Documents folder, so it learns from my writing as well as my speaking. That is way cool, though I have no idea how that helps.




This is a new experiment with Dragon and speech recognition I've got audacity running and I'm not going to try to put punctuation in while I'm talking I'm just going to talk let's see how these two programs record and see whether in fact I can get the audio recorded and the text recorded to I am trying to cause at a fairly regular intervals so the text can catch up but I am not otherwise trying to put punctuation into what I'm saying so I think this sounds fairly natural and then the question is does it actually produce them good enough transcription that somebody else could go through after words and make edits so that it's more readable if you can produce something like this pretty easily we might go fairly far down the path to the ideal that I wanted to achieve and I mentioned in my last blog post that is of making completely good transcripts the first time through that is not possible but what one might produce is an edible edible document it likes edible rather than edit a bowl that one can then actually delivered to somebody else so I must stop now and deliver the raw unedited Dragon documents for you to see

5 comments:

baseball momma said...

hey

baseball momma said...

I am a medical transcriptionist and want to use this for typing in a template in word. I will be listening to dictation through something called express scribe. I will then say what I hear. I am hoping to move around in the template and say what I hear and then correct anything that it types incorrectly as I go along. Do you think I will be able to do this? Do you think it will allow me to listen to dictation and then dictate to it what I want it to type? Any feedback you have would be very helpful and very appreciated.

Lanny Arvan said...

What you are talking about sounds pretty clunky. There is a Medical version of the Dragon software.

Harley Jackman said...

Are there any production transcripts for this sort of thing?

Lanny Arvan said...

I've got a bunch of screen capture videos for microeconomics. They are all captioned. You can see the videos on my ProfArvan YouTube Channel. (The more recent videos are from a different course and are not yet captioned.) All the associated transcript files are available. Note that they are edited substantially from the raw product done with Dragon.