I'd just like to point out that lots of audio recordings kept digitally linked with the transcriptions in a formats that links them through time can be pulled to form training and validation data for developing transcription service software.
I don't think a word document format is what is ideally wanted for the transcribed data, maybe there is a specific XML Schema for that...
Paul (copied in) might know about this as he and colleagues developed a Digital Replay System (http://thedrs.sourceforge.net/) which I think involved some transcriptions from audio among other things.
I don't have direct experience of transcribing interviews or anything else. Anyway, I feel sure there must be some on-going academic research on this.
Andy
http://www.geog.leeds.ac.uk/people/a.turner/
-----Original Message-----
From: Research Data Management discussion list [mailto:[log in to unmask]] On Behalf Of Tim Banks
Sent: 24 September 2013 17:40
To: [log in to unmask]
Subject: Re: Voice Recognition Software for transcribing interviews
In my experience, the reason 'why voice recognition software is not more commonly used' is simply down to the capability of the software currently available. As you say, voice recognition software (Dragon Naturally Speaking Pro being one of the more popular options) has to learn your voice and even then you have to speak very clearly and deliberately, ideally with short gaps between each word, in order for it to work with a high degree of accuracy (& let's face it, when transcribing interviews, accuracy is very important). Apple's Siri service uses some of the most advanced voice recognition technology currently available, running on hugely powerful server farms but still falls very short of being able to accurately transcribe a typical interview.
Re-speaking an interview is the best compromise between speed and accuracy and I suspect this will be the case for the foreseeable future.
Tim
------------------------
Faculty IT Manager
Faculties of PVAC & ESSL
University of Leeds
Leeds
LS2 9JT
-----Original Message-----
From: Research Data Management discussion list [mailto:[log in to unmask]] On Behalf Of Howe, Veronica
Sent: 24 September 2013 17:08
To: [log in to unmask]
Subject: Voice Recognition Software for transcribing interviews
A great deal of research data consists of transcripts of interviews recorded using digital voice recorders and then transcribed, typically into MS Word, by someone listening to it in real time and typing what they hear. Converting digital audio files into digital text files in this way seems quite labour intensive and therefore costly, which makes me wonder why voice recogniton software is not more commonly used.
I gather that the main problem with voice recognition software is that it is designed to recognise a single voice, rather than the two or more voices in a typical interview situation. I understand that some researchers have overcome this by playing back the interview audio so the (recognised) interviewer can repeat what the interviewee said, which is presumably quicker than transcribing it, even allowing for correcting the inevitable inaccuracies.
Given that the interview is such a common research methodology, I would be interested to know if anyone has any experience of using voice recognition software, or using products that can do this to sufficiently high standards.
Veronica Howe
Research Data Manager
Archives and Information Management
Library Services
Kings College London
Room 102
26-29 Drury Lane
London WC2B 5RL
0207 848 1030
Email: [log in to unmask]
Web: www.kcl.ac.uk/library/using/info-management/index.aspx
Connect with us at twitter.com/kingslibraries
|