Neural Audio Technology is used in XM Satellite broadcasts just before the HE-AAC encoder. It improves the HE-AAC efficiency by up to 30%. Read their statement below:

"designed to transparently improve the performance of any audio codec for any transmission system. By correcting anomalies in audio content before audio enters the codec, UltraLink can improve audio codec efficiency by up to 30%. By inaudibly correcting temporal alignment, intensity, and coherence attributes, UltraLink is able to decrease audio irrelevancies at the same time increasing redundancies thus making content easier to encode with less bits consumed. These savings directly translate into increased audio quality, increased bandwidth and less artifacts maximizing any codec’s potential. This correction is accomplished without effecting the audio’s dynamics or spectral shape allowing the user to transparently add the UltraLink to any existing audio chain."

They also have a good white paper about this technology.

My question is: If this technology improves any codec's efficiency by up to 30% without sacrificing anything, why the MPEG is not thinking in a new audio tool based in this technology?

Is there any sofware initiative based on this philosophy?

Regards,
Oki

Did they detect the "transparency" of these manipulations by looking at spectral analyses?

- Lyx

I wdneor if it wokrs on a samilir pcpliinre to the one taht camlis you can utnaendrsd wdros eevn if the mlddie lteters are srcmblead?

Right, right. So it's too bad that they didn't talk about transparency. But it's understandable if this is developed for XM radio, where the point is that it sounds good and not full of artifacts, not that it sounds exactly the same as the original. That said, if this works and is basically transparent, it might be worth working into some codecs or even doing pre-processing. Of course we all know that some music is easier to encode, but today I encoded a cd of pretty-good-sounding rock with -V2 --vbr-new, and the average bitrate is 155 - even on the crunchy songs. So it's less complex where Lame is concerned, but doesn't sound less complex. Anyway, the way that they present this technology might deserve to be made fun of, but it's possible that the technology is good.
Of course, the semester at UNC is starting, and I see all of my bright-eyed students first on Wednesday. So if someone else has free time and is willing to look more into this, it might be worthwhile...

Isn't it just a psychoacoustics engine to pre-process waveform files? Isn't that useless, in itself?

I don't think we need to rework the LAME or Vorbis psychoacoustics engine that much....

(but hey, you never know)

I see ths technology as a Pre-processor for the psychoacoustics engine since it does not remove non perceptible sounds but moves it to a better place for a
psychoacoustics engine.

If HE-AAC can improve up to 30% and XM Satellite Radio is using it showing a clean sound at 64kbps then it is not a joke and maybe LAME and Vorbis could improve their efficency roo.

I would like to know what are they exactly doing with the signal, since it does not have a negative impact on matrixed surround broadcasts but a big positive pump.


Your message has little entropy and a lot of redundancy, as any writting, and even with my limited english knowledge, it can be understood: "I wonder if it works on a similar principle to the one that claims you can understand words even if the middle letters are scrambled?". As you can see the principle is far from being wrong and maybe this preprocessor is doing something similar.

Regards
Oki

Lack of stereo seperation may cause this. Some Portishead music is almost mono, and one track in particular encodes to 99 kbps with Musepack --standard.

Wouldn't this mean, not that file size would be smaller, but that the quality (if it could be measured quantitatively) would be 30% higher at a given bitrate?

Please forgive, a bit of a nub on this placomkex stuff ;-)