Interesting paper. I've always wondered if such SLMs truely improver accuracy or just bias the app to the more frequently requested items. If a test set is...
There are more factors than frequency and weighting, yes? Accuracy is one aspect up for trade-off among many. Latency is another very important consideration....
... accuracy - Yes they do. ... - Yes it certainly does bias, that is exactly how the higher accuracy is achieved. ... why system results will look better, -...
This is another example of real data vs fake data... The majority of VoiceMail messages are much more mundane and predictable than the fabricated ones here... ...
PN, Maybe there was a text reco problem on your end. ;-) The second paragraph pretty much puts the entire article into the proper light. "The aim here was not...
Guilty. I didn't take much notice of that part. My bad. I still get the feeling the article, and most people who responded, expected it to work on that stuff. ...
Couldn't disagree with you more. Your analogy is completely off. How would someone leaving a voice-mail have the remotest idea of the "design parameters" of a...
Phil: The expectation is no less valid, but the use case is far less common. We're talking about inherently buggy software, not has-to-work-every- time jumbo...
The "leaver" of the message is not the user here. I place no assumption of anything on them. The person who signed up to get their received voice messages...
PN, You're right that I should have included the "owner" of the voice mailbox. So, my question corrected: "How would someone receiving or leaving a voice-mail...
... Hi Peter, This may be true, I'll have to take your word for it, but I would certainly agree that these techniques can lead to a greater percentage of calls...
Everything you say is true. Like I said, you have to consider what you are trying to do, and it is critical to consider how important it is to you to be able...
"made complete sense" is completely irrelevant. The "design parameters" are simple - It's designed to work on real voice-mails. These tests were not real...
P.S. When you say "but I see it as managing errors rather than reducing them" - I see it as the opposite. An SLM's main aim is to reduce the absolute error...
That is one of the funniest things anyone has declared to/about me in a long time. Never mind, Peter. You go on standing up for those systems. They need all ...
I know! Thus my surprise. I'm not "pro bad systems", but I am very much against "bad testing". I know you love to turn people's arguments back on themselves,...
Peter, Your arguments make little sense. For the record, I know exactly how SLMs operate, having worked side-by-side with some of the best speech scientists in...
We'll agree to differ on our views here. I would be perfectly happy for my voice-mail transcriber to fail dismally on passages from Ulysses, and I expect...
All the papers from last year's AVIxD workshop have been published on the website. Check them out and let us know what you think. http://avixd.org/pub.html ...
Hi, I am running a simple survey on attitudes to outbound contact and am desperate for non-UK opinions. If any of you have any friends or family who might be...
Quiet list this week. Everyone on vacation? Anyone designing anything cool? If you're going to SpeechTek, what sessions are you interested in? ph -- ...
_The June 2009 issue of ASRNews is on-line_ (http://www.asrnews.com/backissue/validateuser.asp) Apple is sure leaving a lot of opportunities for 3rd party...
This might be of interest to some of you. On August 27 at SpeechTEK, I'm giving a "SpeechTEK University" course on Advanced Speech Application Tuning Topics....
I have been reading a lot of technical and research papers lately, but I wanted to ask some questions of those in the trenches... What are your feelings or...
Title: Voice User Interface Designer Versay Solutions is looking for a Voice User Interface Designer who will help us continue to evolve our industry-leading...
Interesting topic! A few years ago I helped design a financial system that had two distinct sets of caller bases: frequent callers (daily) and infrequent (once...