A Conversation for Speech recognition

A781463 - Speech recognition

Post 21

xyroth

There has been comment on my mentioning the use of headers.

if you read post 2, you will see that I do say it is a relatively trivial fault.

secondly, you will notice that it doesn't mention using guideML headers, only using headers. this can be as simple as a sentance seperating paragraphs that act as an introduction to the content of the paragraph block.

I also feel qualified to talk about the technical aspects, as I have quite a good background in computer based language research in general and speech to text in particular.

I am particularly up to date with the methodologies used in dragon's naturally speaking, and ibm's viavoice. I also know quite a bit about the earlier hearsay program as well.

As the author claimed lack of knowledge, and was making some claims that the knowledge disputed, I thought I out to clarify what he was getting wrong and why.

The only possible thing in that post that I can see for spelugx to get irritated over is the comment about over-hypers being either liars or fools. If it was something else then please drop by my personal space and leave a note and we can discuss it without contaminating this forum.

As to the article, I look forward to reading the rewrite, and will be interested to see which points he accepts, and where he disaggrees with me.


A781463 - Speech recognition

Post 22

Bluttsuuft

To Xyroth,

We're generating some sparks. Wonderful. Get the debate going.

I would also like to clarify a position or two.

Headers are not my main concern. They are sexy and refreshing but I know more than a few beers that also fit that description.

The request for peer review is precisely to provoke a reaction by other members of this august gathering and see what their take on the matter is. I'm not going to start questioning anyones qualifications. Except maybe those of the 43d president of the united states but see, that would be a different discussion, we don't need to go there smiley - winkeye.

I said that I don't know the science of speech recognition. Bigrams and trigrams, how to build a Language Model, your phoneme looks like my sister and what do you mean COM returned an unexpected error code ?
I _do_ have a lot of experience with how the user perceives speech recognition systems. Most notably the Dragon NaturallySpeaking and Voice Xpress family of products and a brief and unfortunate encounter with FreeSpeech 98 by Phillips, a horrible product, not unlike a bucket of fetid dingos kidneys (READ THE BOOKS ALREADY !).

It is in this light that I would like to offer some comments and suggestions. The rewrite should be available soon. You will have something to say about it, I promise smiley - smiley.
I invite anyone to react as irritated and annoyed as they deem necessary. I will be spectacularly unimpressed. This is supposed to be peer review, right ? Big egos do not a great discussion make.

Watch this space.


A781463 - Speech recognition

Post 23

xyroth

looking forward to seeing it (and probably commenting on it as well).

The only reason I posted my previous post was because I seemed to be coming into some stick from doing what I understand peer review (and similar forums) to be about.

If you wish to look up more detailed explanation on the basis for current speech recognition, they are ALL based on a modified form of markov chains, which has to do with conditional probability.

You should be able to look them up in an encyclopedia, or on google.


A781463 - Speech recognition

Post 24

Martin Harper

*rushes in*

I work for a speech recognition company, and I don't understand how the software *really* works. But I do have a little experience.

Regarding the software adapting to you versus you adapting to the software. Sorry, but *both* happen. I know because we run a system where the speaker is completely unaware that they are the victims of speech recognition smiley - laugh, and we get about twice as many errors as a result. That said, some users do "over-adapt", and you rightly point out the dangers inherent in this.

It'd make sense to retitle the entry to something like "Using Speech Recognition Software" - I was initially expecting more of an overview than this 'how to' entry.

Oh, and one more thing:

* Where you have a list of items, use stars to indicate each item
* Like this
* Or this, maybe

GuideML is a lot of hassle - and putting in a few stars like this is almost as good, IMO. But then, maybe I've spent too much time on usenet... smiley - shrug

-Martin


A781463 - Speech recognition

Post 25

The GR Manoeuvre --- a posting a day keeps the reaper away

So, what's the news on this? Bluttsuuff's last post on his Personal Space was 5 weeks ago... so should this be moved back to the Entry?

Caper Plipsmiley - runsmiley - football


A781463 - Speech recognition

Post 26

Monsignore Pizzafunghi Bosselese

To the Flea Market, I'd say.


Key: Complain about this post

Write an Entry

"The Hitchhiker's Guide to the Galaxy is a wholly remarkable book. It has been compiled and recompiled many times and under many different editorships. It contains contributions from countless numbers of travellers and researchers."

Write an entry
Read more