W3C home > Mailing lists > Public > public-audio@w3.org > January to March 2012

Re: fftSize requirement for RealtimeAnalyserNode

From: Raymond Toy <rtoy@google.com>
Date: Thu, 22 Mar 2012 14:27:43 -0700
Message-ID: <CAE3TgXEQ41qrq7z8kDxNop--fP2BVmaBacq24EJ7jPzXMw4q2w@mail.gmail.com>
To: Jussi Kalliokoski <jussi.kalliokoski@gmail.com>
Cc: "Wei, James" <james.wei@intel.com>, "public-audio@w3.org" <public-audio@w3.org>
On Thu, Mar 22, 2012 at 4:07 AM, Jussi Kalliokoski <
jussi.kalliokoski@gmail.com> wrote:

> I hadn't actually even noticed this. Interesting. Why exactly should the
> value be a power of two? It seems to me like the spec shouldn't say
> something like this. Sure, power of two FFTs are easier to implement than
> others, but IIRC they are rarely the fastest. And if we aim for
> performance, the implementation should probably be using something like
> FFTW that allows for any buffer size anyway.
>
>
My recollection is that the power of two FFT is faster (possibly using
radix-4 and radix-8 implementations internally).

Do you have a specific need for a non-power of two analysis?  I'm just
curious.

Also, FFTW uses the GPL license, so if the browser is also not GPL, there
are potential legal issues.  This is where you go ask a lawyer.

Ray
Received on Thursday, 22 March 2012 21:28:12 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Thursday, 22 March 2012 21:28:17 GMT