hiro wrote:According to this site oss production quality creates distortions at about -110db.
Have you tried the test suite, or you took it for granted?
Yeah, it says grc3 is fixed point.
According to this site oss production quality creates distortions at about -110db. If you have a good system and ear you should hear it. But only if you listen to very quiet 24bit files!
Try out this nice internet radio station: http://radio.cesnet.cz:8000/cro-d-dur.flac
libsamplerate has a very linear phase and distortions under 150db, which is not even inside the dynamic range of 24bit.
I'll consider using Z on my ARM, which is very slow with floating points. Thanks for bringing this up again.
If we would not try and discuss the things in a constructive way, nobody would tell us anything new, and we would be here like in the former Soviet Union behind the Iron Curtain.
You seem to refuse to try Petrov's converter, simply because you have not found any reference to it in the net. Yes, it was published exclusively on the OSS4 forum. Would it be better for the OSS4 community, if it would be published on a Russian forum, or simply discussed among a closed circle of friends? If you continue in this way, nobody would share with us any valuable technical information.
That FreeBSD magic tool is of great importance for us. It contains the test suite which allows to study the filter and other things. The filter seems to be the weakest in the old algorithm, and it might be very interesting to learn how they deal with the problemhttp://en.wikipedia.org/wiki/Nyquist%E2 ... iderations
In short, the test suite is of value, and the FreeBSD resampler is, no doubt, an excellent piece of work, which deserves a try, and which deserves respect. But the tests, which were designed for the purpose of proving that it is so good, are likely to be fundamentally wrong. I am very intrigued to study the tests in detail.To summarize:
the so-called "FreeBSD bandlimited sinc interpolator" and the magic filter surely deserve to be studied in detail, even if it may contradict your "mathematical knowledge" and/or "common sense". The latter might be easily dismissed as "superstitions", but there is a certain technical information, which may provide some creative ideas for "clear-cut empirical tests", for instance, this one:
in practice, a signal can never be perfectly bandlimited, since ideal "brick-wall" filters cannot be realized. All practical filters can only attenuate frequencies outside a certain range, not remove them entirely. In addition to this, a "time-limited" signal can never be bandlimited. This means that even if an ideal reconstruction could be made, the reconstructed signal would not be exactly the original signal. The error that corresponds to the failure of bandlimitation is referred to as aliasing. http://en.wikipedia.org/wiki/Nyquist%E2 ... iderations
The reason is simple: libsamplerate , grc3-OSS4, and FreeBSD resampler are of the same sort. And the resampler inside your soundcard is likely to be a similar thing. That is why that magic test suite of FreeBSD might be a very useful tool for us. If it is not perfect, it might be, perhaps, improved, or modified in some way.
It might be necessary to note that the same lossy algorithm is applied for digital recording.
The same bandlimiting filters are implemented, and you have the same "failure of bandlimitation"
which is also "referred to as aliasing
". This is clearly explained in Wikipedia:
It can be argued that analog formats retain some inherent advantages over digital formats. The advantages of analog systems are summarised below:
Absence of aliasing distortion
Absence of quantization noise
Behaviour in overload conditions
Unlike digital audio systems, analog systems do not require filters for bandlimiting. These filters act to prevent aliasing distortions in digital equipment. Early digital systems may have suffered from a number of signal degradations related to the use of analog anti-aliasing filters, e.g., time dispersion, nonlinear distortion, temperature dependence of filters etc. (Hawksford 1991:8). http://en.wikipedia.org/wiki/Analog_rec ... advantages
Each numerical value measured at a single instant in time for a single signal is called a sample; samples are measured at a regular periodic rate to record a signal. The accuracy of the conversion process depends on the sampling rate (how often the sound is sampled and a related numerical value is recorded) and the sampling depth, also called the quantization depth (how much information each sample contains, which can also be described as the maximum numerical size of each sampled value). http://en.wikipedia.org/wiki/Analog_rec ... _recording
The problem is that the lossy sampling algorithm (which is implemented in Linux for everything, for which it should not be applied) does not satisfy the conditions of the Nyquist theorem
The lossy sampling algorithm of Linux and of FreeBSD is fundamentally wrong simply because it does not satisfy the conditions of the Nyquist–Shannon sampling theorem.
The [Nyquist] theorem also leads to a formula for reconstruction of the original signal. The constructive proof of the theorem leads to an understanding of the aliasing that can occur when a sampling system does not satisfy the conditions of the [Nyquist] theorem. http://en.wikipedia.org/wiki/Nyquist%E2 ... ng_theorem
The "true secret" of digital audio recoding, the art of "the real voodoo", and the mythological interpretation of the Nyquist theory might be learned here:
Now we get into the real voodoo. Audiophiles have claimed since the beginning of digital audio that vinyl records on an analog system sound better than digital audio. Indeed, you can find evidence that analog recording and playback equipment can be measured up to 50khz, over twice our threshold of hearing. Here's the great mystery. The theory is that audio energy, even though we don't hear it, exists as has an effect on the lower frequencies we do hear. Back to the Nyquist theory, a 96khz sample rate will translate into potential audio output at 48khz, not too far from the finest analog sound reproduction. This leads one to surmise that the same principle is at work. The audio is improved in a threshold we cannot perceive and it makes what we can hear "better". Like I said, it's voodoo. http://www.tweakheadz.com/16_vs_24_bit_audio.htm
The practical purpose of such mythological interpretations of the Nyquist theory seems to be a pseudo-scientific justification of the fundamentally wrong technologies of digital recording (which are implemented in Linux and FreeBSD). It may help the true believers to fool themselves.
The Linux technologies of digital recording produce ersatz sound, because the lossy algorithm is applied, and the sound is filtered with bandlimiting filters. Then this ersatz sound is processed once more with the same lossy algorithm and the same filters during playback/resampling. As a result, you have what you have, and you are allowed to believe that it is "bit perfect".
"Digital" does contain more information than "analog", but this information is a digital crap produced by the lossy sampling algorithm (that lossy algorithm of digitization is in ADC, which belongs to hardware http://en.wikipedia.org/wiki/Analog-to- ... Processing
). In particular, the standard Linux technology of vinyl digitization produces ersatz sound which tends to be named "crap". It seems to be a kind of stupid business. If you really want to digitize vinyl, you have to make your own recorder which has an exact algorithm of digitization inside.
Digital sound is ersatz sound by definition. This is clearly explained in Wikipedia:
The current crop of AD [analog-to-digital] converters utilized in music can sample at rates up to 192 kilohertz. High bandwidth headroom allows the use of cheaper or faster anti-aliasing filters of less severe filtering slopes. The proponents of oversampling assert that such shallower anti-aliasing filters produce less deleterious effects on sound quality
, exactly because of their gentler slopes. Others prefer entirely filterless AD conversion, arguing that aliasing is less detrimental to sound perception than pre-conversion brickwall filtering
.http://en.wikipedia.org/wiki/Analog-to- ... _recording
Hence, whether anti-aliasing filters were implemented, or not, digital sound is crap in any case. The problems is that "you listen to crap played through crap". If you listen to crap played through the exact resampler, it may sound much better. This should be obvious. If not, try to decode/encode the same mp3 file several times and you may notice dramatic reduction in quality with each re-encoding.
There is already a magic plugin for the Russian "Ultimate Music Player For GNU/Linux", and Russian audiophiles can now enjoy digital sound played through the exact scientific resampler.