VoxForge
VoxForge
Could we please have a prompt saying
"Hum! I don't think so!" and/or "Hum! I don't believe this!"
I try to discriminate between different types of hums.
There are fillers, such as
- "I think it's rather... hum... unusual"
When someone just hums to indicate he/she is looking in his mind for the next word, it means "wait, I haven't finished speaking".
There are agreements
- "You must remember when we were in Paris, this hotel
- Humhum
- it was so filthy"
Here the hum means "yes, I remember"
There is about the same thing but more neutral, a hum which just means "I am listening, go on".
There are disagreements
- "Hotels in Paris are always filthy
- Hum?! I don't think so"
Here the hum means "What are you saying?!"
These hums are not words per se but they are part of the way in which we express ourselves orally, they have semantical content. So to process natural languages, attempting to attach meaning to utterances, they have to be taken into account.
I use short samples (a few second in length) of sound which contain a hum and I try to classify between the different types. The ones I most need are hums expressing disagreement, because they are harder to find in other audio sources (eg interviews).
Mariane
--- (Edited on 9/7/2009 3:21 pm [GMT-0500] by Visitor) ---
Hi Marian,
>I try to discriminate between different types of hums.
Please note that we are only trying to create good acoustic models for simple speech command and control applications. I don't think that this level of sophistication is required for what we are trying to accomplish...
>The ones I most need are hums expressing disagreement,
>because they are harder to find in other audio sources (eg
>interviews).
Why do you need these? for what type of research or application?
Ken
--- (Edited on 9/9/2009 10:37 pm [GMT-0400] by kmaclean) ---
Interesting research, I'm curious to hear whether it is indeed possible to accurately distinguish between different hums. Of course, it will take a while before you get any results.
It might not be handy to collect these prompts/recordings together with all the rest of them, but wouldn't it be possible to separately collect these? Technically it should of course be possible to set up a separate page with another instance of the speech submission app. This instance could use prompts provided by Mariane and she could provide instructions for her subjects on another web page.
This could be set up separately from the rest of the website. Then Mariane could approach participants and do the sound processing etc.
In the (extremely distant) future, such speech might even be interesting to transcribe spontaneous speech. I'm sure that people hum a lot when they're not dictating. We would not create any problems by incorporating such speech into the corpus, since we would keep it separately for the time being.
Well, that's easy for me to say, since it's not me, but Ken who does all the work on the website. And I know for a fact there is a lot of work involved with this at the moment.
--- (Edited on 9/12/2009 6:35 pm [GMT-0500] by Robin) ---