PGGB and HQPlayer Discussion

[LowOrbit]

MLL

 

I liked your post (be rude not to) but you're only fueling my impatience to get my hands on the finished release of PGGBDSD. I heard a few test tracks but without the EQ which would have made evaluation quite a bit more useful to Zaphod and to me. 

 

(Good to know there's more than one PGGB fan in the UK too!)

[LowOrbit]

52 minutes ago, Miska said:

 

Transients in PGGB:

 

 

Same transients with poly-sinc-gauss-long:

 

 

In particular, the Gaussian filters give the most correct transient response.

 

P.S. Your conclusions in your article are pulled out of your sleeve and have so many errors and misconceptions that it is not even worth trying to correct those. For example it doesn't take into account that the original samples of clipped waveform are wrong to begin with, and need correction. Also for example any samples containing aliasing from the source are wrong to begin with. I'm more interested in reconstructing the signal that once entered the ADC, without the ADC introduced errors. If you want to preserve original samples, HQPlayer offers number of halfband-filters, which don't touch the original samples, but also don't correct any of those errors contained there.

P.P.S. But props for you demonstrating massive noise floor modulation in PGGB on your pages!

 

You just don't seem to be able to not pollute this thread with unnecessary trash, @MIska. There was another thread for this kind of discussion.

 

My only question (apart from the obvious) is this: If PGGB is so much worse than your Poly Sinc Gauss filter, why does it sound sooo much better?

[LowOrbit]

I really don't want to get into this needlessly divisive discussion, but here's my take on the whole transient thing.

 

A transient is a short-lived injection of energy but contains very little "information". I strongly disagree with Rob Watts' characterisations of transients because what he describes actually sounds more like the "attack" portion of a musical note (the "A" in ADSR, as referenced above by @kennyb123). (Rob seems quite talented at presenting as fact his views on many aspects of our auditory system without every citing any academic substantiation...) Because of the way our brains perform pattern matching (constantly iterating what it senses from our surroundings) our hearing has evolved to very quickly understand a sound entering our perception, characterising it and acting (in the case of music, no threat (mostly) and moves on). Pitch, spacial positioning and timbre are useful in this context.

 

Transients are by definition too short to provide this kind of information (a transient or impulse really on has an amplitude). 

 

Transients are generated by some musical instruments - that little click when you press an organ key, a strong tap on a piano, percussive instruments generally, a powerful bowing on a string etc - but that's not the same as a simply strong attack portion of a musical note - a staccato note on a trumpet, for instance.

 

In the context of digital recording/replay of music, actual transients will have passed through filters at various stages of the process. In an ideal scenario bandpass filtering at the A-to-D stage would remove the out of band energy. But such brick walling has some strong negative effects. However, some degree of filtering will have occurred prior to reconstruction, which will have removed an amount of the out of band energy.

 

The question then is, how much real impact the behaviour of these filtered transients have when it comes to reconstruction. Artificial transients for the sake of filter behaviour analysis really don't represent what's present in the musical data and in the context of reconstructing a musical event. Several people on this discussion (and elsewhere) claim great prowess at determining the actual impact of filters of different design - I am getting on a bit and can't claim golden ears, but colour me sceptical.

 

 

 

 

[LowOrbit]

I have to agree with @austinpop. Everyone here has huge respect for what HQPlayer has brought to our tiny, almost fetishistic corner of audio and the quality on offer with HQPlayer 5 (and PGGB) are testament to the many years of passionate effort @Miska has invested in showing that separating the reconstruction filter from the D-A engineering piece is a beneficial approach to getting quality audio out of digital.

 

Certainly I would think that HQP at this stage in its product cycle is hugely more widely used than PGGB. And probably everyone who has a PGGB license has HQP too. I use it for streaming and replay all the time.

 

The opportunity to discuss and explore the relative benefits of the different approaches without the entrenched and defensive attitude on display by some parties would be illuminating and beneficial to us all (I say "all" but really we are a tiny band of individuals all stuck at the bottom of a conceptual well, seemingly resenting each other for stealing our oxygen!).

 

Hey ho - as Chris' illustration depicts - the internet runs on this misalignment of views.