Subject: New WWW site: Demonstrations of speech-fluency aids from UCL From: Dan Ellis <dpwe(at)ICSI.BERKELEY.EDU> Date: Thu, 17 Oct 1996 11:23:34 PDTDear List - I was sent the enclosed announcement of a very interesting web-based demonstration of speech fluency processing created by Peter Howell and the UCL Psychology Speech Group. As they claim, it may be the first time that automatic processing of submitted files has been offered on the web - an exciting trend. DAn. ------- Forwarded Message Date: Thu, 17 Oct 1996 13:20:39 +0000 From: stevie(at)psychol.ucl.ac.uk (Stevie Herbert Sackin) Subject: new WWW site Dear all, We are emailing all the people we have addresses for to announce the establishment of a new website that demonstrates processing of speech data. The address is: http://www.speech.psychol.ucl.ac.uk/index.html One feature of the site is that it allows anyone who is interested to submit their own speech data for processing. There are a few limitations which affect who will be able to use it: 1. Speech file format restictions (details are given on the site). 2. To access all facilities needs a browser which can handle frames. 3. You need a sound card if you want to hear the speech examples. 4. Uploading your files for processing takes a bit of time so be patient and please lay off the keys while you're waiting - we will fix this problem shortly. We intend extending what is on offer as soon as we can, both in terms of processing longer files and by offering additional forms of processing. At present, the only processing offered is syllabification. There is an example where we have marked all vowels with a click and you can hear the speech and superimposed clicks in one of the demos. Sample data you may wish to submit can be processed in this way and results accessed at the site. You can also hear a demonstration of one of our early attempts at automatically marking out stuttered sections of a child's speech. For the technically minded, what we compute are the words spoken by the child (using DTW) and then apply an ANN classifier to each word. In the example you hear the speech with a tone superimposed on the bits that are thought to be dysfluent. We are in the process of updating the software to use the new segment markers and applying a better ANN classifier so if you are interested watch this space. When that is released we will try and set it up so you can try it for yourself (there's no point in letting you use the current system at present as the new one's much better). We are interested in comments, requests and so on concerning how we can improve things so let us know what you think (please fill in the questionnaire when you visit the site). Please pass this information on to any friends and colleagues you think might also be interested. Incidentally, we have searched around (date 10/10/96) and can't find any other site that offers online Automatic Speech analysis facilities over the web (we found a few where you can ftp data across for processing but that's not the same thing at all). Granted, we're only offering automatic syllabification at present but even so, we'd like to claim a world record on this. If there is anyone out there who knows that we are wrong, then perhaps they could let us know. Finally, we would like to mention the Wellcome Trust's role in this endeavour: Without their support, the development of the software and opportunity of making this available on the website would not have been possible. In the spirit of international co-operation in our mutual endeavour in understanding the nature of speech, particularly those concerned with communication breakdown in children, regards to all. Peter Howell, Kazan Glenn, Stevie Sackin and James Au-Yeung. ------- End of Forwarded Message