Slashdot Mirror

LeelaZero - an open source go bot, has beat 9d professionals and other lower ranked professionals. It is also ranked #3 in the world in gobot competitions, and that was with using half or less of the hardware resources that many of hte competitors had (LeelaZero was using 4 1080 TI GPUs; the competitors had 10 1080 TI GPUs).

It still hasn't reached the level of AlphaZero, but if you'd like to help it do so, you can contribute here.

http://zero.sjeng.org/

Note that they benchmarked against LeelaZero, but had it misconfigured - they gave their bot 80,000 playouts, and LeelaZero 50 seconds per move, but left a default where LeelaZero doesn't use all of its time. So often it was moving in 3 seconds. It might well be weaker than LeelaZero on similar hardware when LeelaZero is correctly configured.

It can really pay off to do some optimizing of your userChrome.css file. For example, one thing that annoyed me greatly was that the tabs don't move next to the menu, losing precious vertical screen estate. And most of the context menu entries just get in the way. All of this is fixable without any plugins or whatever. There are many examples of annotated userChromes around, if you're anything beyond a casual user it pays off to have a quick look.

With a good enough algorithm most people, including those with well-trained ears, will not be able to consciously distinguish the two sounds. But that does not mean that these people don't subconsciously react differently to them. One way to measure that might be to measure brain activity in various regions of the brain, which is exactly what this article mentions. The problem is that that type of test is always going to show a different reaction which is something the makers and users of audio codecs often don't want to hear.

The major problem here is what does the brain activity data mean? Even if you can see a difference in brain activity for a 16 bit/44 kHz PCM file verses a 128 kbit/sec VBR AAC file how do you determine if one format is preferred over the other?

You end up still falling back on subjective measures, it's much simpler to have a large number of participants and then ask them questions like, "Which recording did you prefer?" The data from a properly run survey is much more likely to yield meaningful conclusions than scans of brain activity. We are, after all, dealing with music - a highly subjective art form.

One notable feature of DSD is that dynamic compression occurs at higher frequencies yet the frequencies are able to be reproduced accurately. Contrast this with PCM where the dynamic range is fixed (i.e. 16-bit, 20-bit, 24-bit) but at higher frequencies the tonality is not as pure because it's impossible to represent anything other than a square wave at the nyquist frequency which is exactly 1/2 the sampling rate. Of course, a filter is applied to make that into a more pleasant sine wave. Now consider a frequency that is not exactly 22.05 kHz but perhaps a little shy of that. It's almost impossible to represent this accurately with PCM. The result is that you actually get a slightly oscillating frequency somewhere around the original frequency.

What you are describing is a phenomenon known as aliasing.

I'm not sure you completely understand how the Nyquist-Shannon Sampling Theorem works. It boils down to the fact that as long as you sample at a rate greater than double the maximum frequency you want to capture, you will get no aliasing. This means that if you sample at 44.1 kHz then all frequencies below 22.05 kHz will be represented accurately. If you sample a frequency just shy of 22.05 kHz you will NOT "get a slightly oscillating frequency somewhere around the original frequency".

It is true that DSD has a variable dynamic response that depends on frequency but that works both for and against DSD since higher frequencies tend to less accurately represented than lower frequencies. In fact there is a lot of discussions (PDF file, see page 8, section 3[c]) that conclude that the current implementations of DSD produce worse quality per bit than an equivalent bit-rate PCM sampling. There are solutions to these problems but they are very complex and involve a mix of DSD and PCM sampling methods, so much so that the line between DSD and PCM blurs considerably.

This has a serious effect on how an album is mastered. When the target format is CD the producer can cause the CD player to output extremely loud high frequency sounds though not particularly accurate frequencies. This is reflected in the current crop of music which is often extremely loud and to many ears just sounds like a bunch of noise. Metallica's self-titled black album was one of the first to use severe dynamic compression to make the album sound super loud. Comparing it with modern CDs we can see that that album was relatively tame.

Again you are mixing up sound levels with frequencies. Severe dynamic compression basically limits the number of sound levels which are utilized,

DSD uses sample rates of the order of 4MHz

<flamebait>Yes, and DSD is not what people in their right mind use ;-)</flamebait>

From Why 1-Bit Sigma-Delta Conversion is Unsuitable for High-Quality Applications :

The manufacturers of high-quality converters struggled mightily to produce 1-bit devices that met the performance goals of the industry. But, they could never eliminate all the undesirable artefacts of such converters, and after more than a decade of trying, they came to the realization that they could produce better performance by using multi-bit converter architectures in their products.

Anyway, you are technically correct in pointint out the flaw in my wording. People use multi-gigasample ADCs when doing things not related to audio. And yes, many audio-frequency ADCs do oversampling internally... but the output is still on the order of 44 to 96 kHz at full-bit width (16/24 bits).

SACD is freaking the best digital sound format on the market AND it's 1 bit (DSD). How cool is that?

not too cool.

Why 1-Bit Sigma-Delta Conversion is Unsuitable for High-Quality Applications

Single-stage, 1-bit sigma-delta converters are in principle imperfectible. We prove this fact. The reason, simply
stated, is that, when properly dithered, they are in constant overload. Prevention of overload allows only partial
dithering to be performed. The consequence is that distortion, limit cycles, instability, and noise modulation can
never be totally avoided. We demonstrate these effects, and using coherent averaging techniques, are able to display
the consequent profusion of nonlinear artefacts which are usually hidden in the noise floor. Recording, editing,
storage, or conversion systems using single-stage, 1-bit sigma-delta modulators, are thus inimical to audio of the
highest quality. In contrast, multi-bit sigma-delta converters, which output linear PCM code, are in principle
infinitely perfectible. (Here, multi-bit refers to at least two bits in the converter.) They can be properly dithered so
as to guarantee the absence of all distortion, limit cycles, and noise modulation. The audio industry is misguided if
it adopts 1-bit sigma-delta conversion as the basis for any high-quality processing, archiving, or distribution format
to replace multi-bit, linear PCM.

Agreed. ID3 is a horribly limited hack (hence the need for ID3v2) on a format (MPEG) that was never meant to support arbitrary user metadata. I have no problem with my Ogg Vorbis music files. Ogg files are designed to hold metadata in a simple but flexible name=value format. Title, artist, album, track number, genre, whatever you want. You can even add your own fields, like Vorbisgain data.

ID3 is just a block of data tacked onto the end of the MPEG I layer 3 audio file. It isn't a part of the MPEG standards. Frankly, it's very poorly designed, with fixed-length fields and no way to add extra fields. Except by adding even more data at the end, hence ID3v2. Bad, bad, bad.

• ReplayGain for MP3s
• ReplayGain for Ogg Vorbis files

Cheers,
Toby Haynes

Check it out!

You won't believe your ears!

Try Sultans of Swing in 197K!

Amazing! This is truly insane!

Apple, you must support Vorbis. If the iPod does Vorbis, I buy one, and that's a promise.

The gentleman who wrote sjeng also wrote prototype Vorbis 1.0 encoder that can go down to bitrates of 4kbps that he claims can give a listenable stereo stream. IMHO that's bigger news than source to Chess.app 2.0.

I'm not familiar with the state of MP3 tools which support ReplayGain, but I know that Gian-Carlo Pascutto just wrote a tool to add ReplayGain information to Ogg Vorbis files. There is an XMMS support in CVS which uses the information, and I just got done adding support for ReplayGain to ogg123 (it will be about a week before it goes into the xiph.org CVS pending the approval of some other changes). Winamp also supports ReplayGain using Peter's Vorbis plugin

There's a batch Ogg replaygain tool at: http://sjeng.org/ftp/vorbis/

ReplayGain tself is explained at: http://www.replaygain.org

The latest XMMS plugin already supports replaygain (as does latest Ogg123), and it should be in the Winamp plugin soon if not already. Right now it's up to individual apps to support ReplayGain, but we're deciding on an easier way to encourage/include support with core Ogg.

Monty