Ghost Update: Demo 4

[xiphmont]

Turns out I missed blogging about the latest Ghost update... back in November...

Ghost Demo4 is up on the demo list showing the sinusoidal extractor doing some very early sinusoidal tracking frame to frame, and a very early example of the analysis performing real sinusoidal/non-sinusoidal audio splitting. Pictures and interactive listening, oh my!

It looks like I'll be putting a month or two into transOgg before getting back to Ghost work (and demo 5). The work that went into demo4 raised a number of questions I'm not sure how to approach answering yet, so I'm going to let that percolate for a bit.

Comments

Fourier

[daniel hendrycks (from livejournal.com)]

Will this affect anything Xiph does with Ghost? http://www.theregister.co.uk/2012/01/19/mit_fast_fourier_transform/

Re: Fourier

[xiphmont.livejournal.com]

Possibly but probably not. It's not a fundamentally new transform, it's an optimization (in special cases) of the existing FFT. There were already faster-than-O(n*lg(n)) special cases of the FFT, this announcement appears to add to that set.

But hey, faster FFTs are always welcome.

Has Ghost been shelved?

[(Anonymous)]

Has Ghost been shelved due to Opus being good enough?

Re: Has Ghost been shelved?

[xiphmont.livejournal.com]

Officially no. However, Opus wasn't expected to be anywhere near the quality of Vorbis, let alone be a viable replacement. The fact that it is has moved Ghost's priority down somewhat.

It's still the case that if Opus can achieve those kinds of gains with low latency, Ghost would be able to do even better-- but there are more pressing things to work on now, such as the transOgg update.

grad students... I need an endless fleet of grad students....