According to the recommended MPC settings, the standard profile averages at 170 kbps
According to this thread, it can average at 176 kbps.

I've calculated the bitrate of the 16 first albums I encoded with MPC 1.15r --quality 5 xlevel.



The average is 181 kbps, but we can see that the most common bitrate is a bit higher because of the asymmetry of the curve.

The two highest bitrates, 193 and 205 kbps are a vinyl copied on cassette, and a vinyl, respectively. All others are CDs. It is possible that the background noise and/or clicks have increased the bitrate. I can't compare with 1.14 since I didn't encode anything with it.

Is it possible that there has been a bitrate increase between 1.14 and 1.15r ?

I think Frank said bitrates raised slightly with the 1.15x versions (only a few kbps), but I'm not sure.

I still see some wildly low bitrates - just encoded Hendrix - Are you experienced? And two songs were around 99 kpbs with -q5!! The rest where 140-180. An explanation of why this happens would be nice.

There's a pretty big increase in bitrate between 1.14 and 1.15r. "Pretty big" means an additional 4-6kbps on average, but that's a lot more than most version progressions. The reason for the increase is the new tonality measurement algorithm in 1.15r. Since it's still not perfect, I would expect another increase for 1.16.

I use MPC for some lower bitrate needs (~128kbps) and I am going to be forced to use 1.14 for these jobs. I think that the extra bits needed by 1.15r aren't a "good value", i.e. I really can't tell the difference between 1.14 and 1.15r under any kind of normal listening. But I understand that Klemm & Co. want to make --standard as transparent as possible and that may very well require more bits. But when you are targetting 128kbps, the extra 6kbps "penalty" caused by 1.15r makes you cut back in other areas, like the noise-masks-tone ratio. So while 1.15r may be a good step forward for transparency, it can also be considered suboptimal.

This is a common observation. If you get 100kbps out of -q5, 99.5% of the time it's due to the source material being mono or very minimally stereo.

The 170-180 kbps average bitrate is a good estimation for some kind of music. Try classical, especially music of the baroque period, and you can increase this estimation up to 200 kbps (maybe over).
One of the higher track I have (harpsichord piece of JP Rameau) is rated at 233 kbps with --standard (1.14) and 249 kbps with the same setting and 1.15r version. 16 kbps difference. This is really big (7%).

Fundamental improvements in the psymodel should not cater to bitrate concerns regarding presets (at least, it shouldn't be a priority until after the psymodel related work is done).

I see no reason to use 1.14 over 1.15r because of the latters bitrate increase. We have an adjustable quality scale that we can use now to lower the bitrate, so there should be no problem really, you just need to readjust your scale to match these new changes.

Pio wrote:

" The two highest bitrates, 193 and 205 kbps are a vinyl copied on cassette, and a vinyl, respectively. All others are CDs. It is possible that the background noise and/or clicks have increased the bitrate. "



Yes, I assume this, too, that background noise/clicks increases bitrates.

I have lately encoded a CD to mpc, which was recorded from old masters from 1930-1939 time period.
1992 or 1994.
Those sound engeneers did not take away the background noise.
It was all mono, but bitrates were as high as low-bitrate stereo mpc's.
But the background noise was perfectly transparent in mpcs...
With other mono-tracks, eg. from latest Rolling stones forty licks CD, which do not contain such annoying abckground noise, I had really low bitrates, about the half of normal stereo mpcs.
So, all those facts indicate clearly, that background noise causes high bitrates.
I encoded a Test-CD, which contained pink/white noise and other test signals, like discrete frequencies or noise located at special speakers, surround etc.
The highest bitrates were clearly those tracks with noise all over the frequency sprectrum (white/pink).

Math is the language of the universe. Everything and anything can be represented by numbers; graph those numbers and patterns emerge.

Quote - PI movie

It's hard "to graph" noise which is by nature more or less random thus "pattern free". I guess excepting 50Hz or 60Hz Hums. Anyway, lossy encoders have a hard time when dealing with noise, be it (moving) image or sound.

Guess I had too much coffee today.

Most noise can be exceptionally well mathematically modelled...
eg. pink noise is fractal noise - theoretically it is possible to express it as a function.

But how to do this in audio encoding?

I guess my point is: if I have to adjust my scale downward to match these new changes, achieving poorer quality overall is entirely possible. A 2-3% bitrate increase may not require a large downward shift in a user's chosen scale points but you are still forced to make a substitution: degrading everything a little in order to accommodate a new change.

I haven't been technically briefed on the new changes in 1.15r but it seems to me that they should kick in for --quality 5+ scale points rather than universally...much like --minSMR.

Well, by definition, truly random noise (pink, white or anything else) cannot be expressed in a smaller form.

Now, you can shrink it (keep just its characteristics) to reproduce nearly equivalent noise later on, but the individual samples will have different values that before.

In other words, you can generate noise with a fractal, but if you take a noise from another source, the fractal formula will be the same size as the noise itself.. so it's not that easy

Yes, but I assume that if I use a quality below 5 for 1.15r in order to accomodate bitrates, the overall quality will be lower than with 1.14 q5, isnt it?

Otherwise, that would mean that quality scales linearly with bitrate and viceversa, no matter what encoder version. I think this is not the case.

Currently, I don't care much about those extreme cases that 1.15r solves, so I prefer to use 1.14 and have a lower bitrate.

BTW, in most pop/rock songs I've tried, the bitrate increase with 1.15r is of around 10 Kbps, being the overall bitrate between 180 and 200 Kbps.

I think perhaps what Bedeox means is that to us, it's all noise. So if you can figure out where the noise is, and characterize it, the individual samples don't matter, because our ears can't resolve it anyhow. Psychoacoustic stuff.

Well.. yeah - I mentioned it btw:

This is a difficult thing to do though. There are many types of noise that all sound different, and each of them can be represented in several ways.. this requires a thorough analysis in both time and freq domains.. it seems statistics alone won't cut it.