Researchers at Stanford University (USA) and the University of Edinburgh (UK) have tested various automatic speech recognition (ASR) systems and found that in general they have a more difficult time recognizing speech from males than females. One of the causes mentioned by the researchers was that the men tended to use fillers such as "emm" more often.
I should point out that the tests were conducted using recordings of telephone calls. This is important because the type of "channel" the audio is being carried over makes a significant difference to a computer-based speech recognition system.
The work was sponsored by the US Office of Naval Research.
For a quick overview of the research, you can read the BBC (center-left news media, UK) article here.
Enjoy!
Showing posts with label speech recognition. Show all posts
Showing posts with label speech recognition. Show all posts
Thursday, March 04, 2010
Wednesday, February 18, 2009
Phonetics: The Sounds of American English
Check out this educational page on the Sounds of American English from the University of Iowa (USA) - complete with cut-away animations of what is happening in the vocal tract and videos of the face during articulation. My wife is probably tired of hearing me point out over the years that this is exactly what is missing from educational software for learning foreign languages. Learning to properly articulate a sound element (phoneme) is so much easier if you know what the articulators (i.e. lips, tongue, jaw) are doing! Well done.
(Hat tip: Good Morning Silicon Valley)
Wednesday, December 31, 2008
Human Auditory System: Speech Intelligibility overview
A colleague pointed me to the following overview of speech intelligibility on the Sound & Video Contractor website. It was written for intelligibility during teleconferencing, but the information in it is directly applicable to audio forensics. Special attention should be paid to how reverberation and signal-to-noise ratio (SNR) affect intelligibility.
Enjoy!
Enjoy!
Sunday, May 11, 2008
Human Auditory System: Voice Risk Analysis
The Economist (UK, center-left economics and news magazine) has a short article in the May 8th, 2008 edition on Voice Risk Analysis, which is another name for Voice Stress Analysis or Micro-Stress Analysis. You might recall from my previous posts (here and here) on this topic that I believe voice stress analysis to be a workable interrogation tool, when correctly administered, but then, so again can the copy machine in the hallway - it all depends on whether the subject is, or can be made to be, concerned about it. It is not, however, a lie detector. This author of the Economist's article seems to have come to a similar conclusion.
An interesting point in the article is that, at least in one real-world deployment of a commercial system, the biggest benefit was gained from simply telling the claimants that such a system would be used on them - which caused many to drop their claims prior to being "analyzed" by the system.
An interesting point in the article is that, at least in one real-world deployment of a commercial system, the biggest benefit was gained from simply telling the claimants that such a system would be used on them - which caused many to drop their claims prior to being "analyzed" by the system.
Friday, December 07, 2007
Speech Recognition: Application to university lectures
Technology Review (an MIT publication) has a comprehensive article on an up-to-date application of speech recognition, including both the recognition engine and the user interface layers - namely, MIT's Lecture Browser website, a search application for video lectures. The article is very well written and is approachable by anyone with a bit of science background.
Labels:
audio,
computer,
filtering,
speech recognition,
video
Subscribe to:
Posts (Atom)
