Stereo Tool

Ok... I'm back to my PC platform here - thus, I re-welcome "Stereo Tool" for my processing solution!

Here are links to my stream:

http://gms.go.dyndns.org:8000/listen.pls for aacPlus V2 48kbps

http://gms.go.dyndns.org:8004/listen.pls for aacPlus V1 32kbps

Please let me know what you think about the sound quality - any artifacts that you're noticing, timbre issues, dynamic range, etc. It plays a mix, with some holiday (Christmas) music in the rotation -- as well as my local Kansas City forecast at the top of the hour, every hour. I haven't made any new station IDs yet, so that is disabled.

The goal here is to normalize and pre-process for artifacts - to effectively reduce bitrate as much as possible without adversely impacting quality. Here is a sample playlist - the last ten items played:
08:35:08 Metallica - The God That Failed [Metallica (The Black Album)]
08:30:38 Disturbed - Indestructible [Indestructible]
08:25:55 Bachman Turner Overdrive - Takin' Care of Business [The Best Classic Rock - Speed Session]
08:20:28 The Who - You Better You Bet [The Ultimate Collection Disc 2]
08:17:02 Reba McEntire - Santa Claus Is Coming Back To Town [Secret Of Giving: A Christmas Collection]
08:12:47 Yes - Owner of a Lonely Heart [The Best Classic Rock - Speed Session]
08:10:16 Alan Jackson - Little Bitty [Greatest Hits, Volume II]
08:06:40 Billy Squier - The Stroke [18 Screamers From The 80's]
08:02:36 Mariah Carey - Emotions [Greatest Hits (Disc 1)]
08:01:52 Weather

Thanks! As always, it is a work in progress.

I am using 3.40 BETA.

I made some substantial processing improvements, concerning some advancement in my knowledge of the Spectral Band Replication (SBR) portion of the CODEC. As always, feedback is appreciated.

Hi,

The quality is good, i listened to the 48kbps stream, good dynamics, the high's are a bit to loud for me but that's a matter of "taste"

Cheers!

Martijn

Thanks! That's the kind of feedback I was wanting to hear!

I'd questioned about those highs myself. It's a work in progress. I just toned them down a bit.

You have to be careful when "tweaking" highs on the aacPlus CODEC. It's easy to fill it with sibilance or slushiness, since the highs are reproduced in the decoding process -- synthesized, effectively.

Ok, I've just tuned in (really didn't have time the last few days because I was trying to get the 3.40 version ready).

For me, highs can never be too loud. In fact, the first thing I do when I get my hands on audio equipment is to turn the treble up as far as possible. And I always leave it there

So for me the amount of highs is really low. But given the fact that you have to deal with SBR compression, that's probably the only way to get a decent sound. (Which might also explain why 2 years ago when I tried out MP3Pro 64 kbit/s which also uses SBR for my own station it sounded so distorted...)

Also, I have the impression that a part of the high frequency spectrum is a lot softer than the rest - I think you referred to it as "watery cymbals" in another post, although it is a constant lack of those frequencies, it's not changing as in low bitrate MP3s. If I feed the sound through Stereo Tool at my end (press RESET, turn Multiband compressor on, Pre Amp to 10 and Post Amp to 0.5), the watery sound disappears completely and - at least in my opinion - the sound becomes much better. But when I look at the output bars of the multiband compressor, I don't see any strange jumps, so it might just be a matter of taste (I guess I just don't like the way the high frequencies dropoff is configured). The fact that others said that there was too much highs seems to confirm that. And I noticed that after listening for a while I'm getting used to the sound.

Having said all this, when I do increase the treble the sound is good, and I'm noticing far less encoding artifacts than I would expect at such low bitrates, even when sending it through Stereo Tool again (which amplifies soft sounds, including artifacts). Which is pretty impressive.

SuperH, I just had an interesting idea - maybe you could check if this would work...

Could you try the following:
- Turn on (if you haven't already) the Stereo filter
- Set the Phase slider to '0'

I'm curious if this leads to less artifacts in your signal. (You might be able to increase the highs more before getting the "watery cymbals" effect.)

Why:
Doing this removes phase differences between the left and right channel. At least when using MP3 compression at Joint Stereo setting (the default for low bitrates), those phase differences are very difficult and can cause huge artifacts.

So: Reducing those phase differences MIGHT reduce the amount of data that the compressor needs. Which would lead to less artifacts. (Using this setting, you can even get fairly good results at 96 kbit/s MP3s).

If that is also the case for AAC, I can create a test version of Stereo Tool for you that removes or decreases the phase difference for the non-SBR part of the signal, but keeps it for the SBR part - or the other way around. That should still SOUND more or less the same as it does now, but could possibly lead to (far?) less artifacts.

Hans,

Thank you for your input! My highs have been dropped since the review about them being too sharp. The average dB level at about 12kHz is 2dB below what it was. I think it sounds more muffled than what it did. I plan to restore to the settings I used prior, and build onto that.

I tweaked a bit more. I'm playing with Phase = 0. It seems to have cleaned up some of the "swishiness".

Normalization works the same as it has before. I'm a bit proud of that algorithm (gloat... gloat...)



Input (Top), Output (Bottom). You can probably figure out the normalization scheme -- while trying to preserve instantaneous dynamic range. The volume at the pre limiter slowly raises as necessary -- and it rapidly drops - by being "tapped" down with each sharp peak. Pretty cool. Pretty typical for radio stations that want to preserve "punch" in music. I never thought it was possible -- I don't need a "hold time" configuration after all!

SuperH: Another question... I've been reading into the SBR encoder a bit, and I'm really wondering what effect the the Loudness filter has on the encoding quality.

Reason: SBR guesses the high frequency spectrum from the low frequencies ( http://www.chiariglione.org/mpeg/techno ... /index.htm ). Loudness changes the relation between frequency bands, and it MIGHT make things easier for the AAC encoder. (It don't know if things are getting easier or more difficult though).

How to test: (Assuming you're not using Loudness now):
- Set Loudness to a high value (for example 3.0)
- Set the Post Amp volume to a lower value (0.33 to match the 3.0, that should give about the same output volume)
- Keep "HARD LIMIT" off! The low Post Amp value will avoid too loud peaks.

If you ARE using Loudness, please try if turning it off makes the sound better or worse.

Note: Please use the 3.40 release version for this, not one of the 3.40 Beta's! Because the betas cause other ill effects that were solved in the release.

What I'm hearing, the loudness filter seems to be smooth enough to have a minimal impact on encoding quality.

It slushes up the percussion a bit -- but not too bad considering the aggressiveness of the algorithm. I'm actually impressed with how well the loudness filter is working. Enabling hard limiting worsens the audio quality, of course, but not as much as I'd think. My algorithm DOES use "Hard Limiting".

Going back to my algorithm cleans it up quite a bit - but it still needs some work. I have heard Neural Audio in action doing this task, as well as the OptiCodec by Orban. I'm guessing your DSP coding is actually better than theirs. Audiophiles and enthusiasts designing stuff tend to make things sound better than "paid engineers".

Nice work on the loudness filter! That's impressive.

Actually, I think the reason isn't that people don't WANT to make a good filter or don't hear the difference, but that the mathematics behind it is getting so complex that they probably don't know how to do it. (In fact, I don't even fully understand how my own filter works, and I designed it... You may have seen the number of in-between versions before I reached the current quality in the News forum!)

About the slushyness: I may still have some tricks up my sleeves to reduce that slightly. The current Loudness filter consists in fact of 6 filters in a row that all have a lot of parameters that I can set. And since releasing 3.40 I have gotten some new ideas that might improve the sound. I'll look into those in a while.

(By the way: If you are getting this slushyness: Do you have FM pre-emphasis turned on? If so, and you're not broadcasting an FM signal (but using it for web radio), turn it off! That should reduce the slushyness greatly, because it is mostly audible when very loud highs are present - and FM pre-emphasis makes the highs a lot (more than 10 dB!) louder.)