From: Fran Barg
Posted: Tue 2/07/2002

I am beginning several large projects that will include interviews conducted by multiple research assistants. I am interested in finding out if anyone has had success using voice recognition software to generate transcripts that then can be imported into Nudist?

From: Elliot Richmond
Posted: Tue 2/07/2002

The short answer is, probably, that what you have in mind won't work.

Voice recognition software must be trained to a particular person's voice.
Some versions allow for multiple users, but each user must be trained separately and the software will only recognize one user at a time.

Also, using even the best recording equipment available will not produce recordings of sufficient quality that they can be fed directly into the software. Speaking "live" into a high quality microphone is about the only technique that produces reasonable results. Even then, mistakes will occur frequently and it is better to correct these mistakes as they happen.

Some software packages will accept input from a high quality digital recording device, but the software still must be trained to recognize the device as a user.

From: Aleksandra Belofastov
Posted: Tue 2/07/2002

I use Dragon Naturally Speaking (version 4 - there is a version 6) to transcribe my interviews directly onto the computer. Dragon requires that you train it to the sound of your voice which would mean that each of your research assistants would need to spend time doing this. The transcription into the computer is saved as a wave file (this can take considerable memory) you then set the computer to type out the transcript and save it in plain text. Unless you have a good deal of memory, I would suggest deleting the wave files at the end of this process. I should also mention that version 4 is not perfect; I have had to 'tidy' my interviews following transcription as Dragon doesn't always understand what I am saying!

Version 6 may be better. Needless to say, having attempted both methods of transcription (i.e. directly typing from the audio cassettes and using Dragon) I have most definitely saved considerable time using the latter method.

Importing into NUDIST is relatively easy from this point. If you are using the older version, you need to make sure that the plain text interview looks the way you want it (i.e. paragraph returns in the right places & so on), the newer versions (I am using #5) allows you to edit the interviews even following importation.

From: Shigeko Izumi
Posted: Tue 2/07/2002

I think the problem of voice recognition program is that it has to be trained. I understand your research assistants spent time for training, but how about the voice of interviewees?

How did you trained the program on the interviewees' voice which is tape recorded ( I guess)?