Re: ADF: "Hearing" a sound programatically? [Core Audio]
Re: ADF: "Hearing" a sound programatically? [Core Audio]
- Subject: Re: ADF: "Hearing" a sound programatically? [Core Audio]
- From: James Maxwell <email@hidden>
- Date: Wed, 16 Oct 2013 17:39:04 -0700
This isn't easy, but certainly possible -- allowing for some degree of error! ;-)
Comparing to an audio file won't be practical. Instead, I'd do some analysis of a few different spectra of the kind of blowing sound you're looking for, then hand-craft some heuristics based on that (for example, it would be mostly inharmonic, and it would probably have a pretty characteristic onset). Which is just to say that, since you know what you're after, you could hand-code some constraints.
From there you could do spectral analysis of the input and see whether it satisfies the constraints. Probably you'd find that even just a "noisy" sound with a high amplitude would be good enough (in fact, maybe even just amplitude would work -- but then people might figure out that they can shout the candles out!). I'm sure there's an open source spectral analysis package you could use somewhere. (For example: http://lists.apple.com/archives/coreaudio-api/2011/Jun/msg00061.html)
Another option, though probably overkill in your case, would be to use something like an HMM to learn it (http://www.di.ubi.pt/~jpaulo/competence/tutorials/hmm-tutorial-1.pdf). You could train that up yourself (still need the spectral analysis stuff, mind you), and maybe with a few friends (to give it some generalization capability), and perhaps even allow it to continue learning from the user. But this would take some work, for sure, and is probably way over the top...
I'd say the first option is easier, since there should be some really clear characteristics you can look for.
J.
On 2013-10-16, at 5:19 PM, Apple Developer Forums <email@hidden> wrote:
> View or Reply online: https://devforums.apple.com/message/907941#907941
>
> --- LaDauphine
>
> I want to write an app that responds to the sound you'd make when blowing out birthday candles. The end user would be making this sound, and I want the app to recognize it, then do something. So my guess is that this isn't really speech recognition but instead parsing and "matching" an audio file to a similar file. What frameworks would I use to do this? Is this even possible? Just want to get pointed in the right direction... thanks.
>
> ---
> You currently receive email updates for all posts to Core Audio
> Unsubscribe at: https://devforums.apple.com/people/jbmaxwell?view=watches
_______________________________________________
Do not post admin requests to the list. They will be ignored.
Coreaudio-api mailing list (email@hidden)
Help/Unsubscribe/Update your Subscription:
This email sent to email@hidden