Myths of Digital Music Part 5

Myth 1

The wider the spectrum, the better the recording (over spectrograms, auCDtect and frequency range)
Today in forums, unfortunately, it is very common to measure the quality of a track with a “ruler in the spectrogram”. Obviously, for the simplicity of this method. But, as practice shows, in reality everything is much more complicated.

And the point is this. The spectrogram visually demonstrates the distribution of signal power at frequencies, but cannot give a complete picture of the sound of the recording, the presence of distortions and compression artifacts in it. That is, in fact, all that can be determined from the spectrogram is the frequency range (and partially, the density of the spectrum in the HF region). That is, in the best case, by analyzing the spectrogram, you can identify the upconversion. The comparison of the spectrograms of the tracks obtained by encoding several encoders with the original is absolutely absurd. Yes, you can identify differences in the spectrum, but determining whether (and to what extent) they will be perceived by the human ear is almost impossible. We must not forget that the task of lossy encoding is to provide a result that the human ear cannot distinguish from the original (not with the naked eye).

The same applies to evaluating the encoding quality by analyzing the output tracks with the auCDtect program (Audiochecker, auCDtect Task Manager, Tau Analyzer, fooCDtect are just shells for the one-of-a-kind auCDtect console program). The auCDtect algorithm also analyzes the frequency range and only allows you to determine (with a certain degree of probability) whether MPEG compression was applied in any of the encoding stages. The algorithm is designed for MP3, so it is easy to “cheat” with the Vorbis, AAC and Musepack codecs, so even if the program writes “100% CDDA”, it does not mean that the encoded audio is 100% identical to the original. .

And going straight back to the specters. Also popular is the desire of some “enthusiasts” to turn off the low pass filter on the LAME encoder at all costs. There is a lack of understanding of coding and psychoacoustic principles. First, the encoder cuts the high frequencies for one purpose: to save data and use it to encode the most audible frequency range. The extended frequency range can be fatal to overall sound quality and cause audible coding artifacts. Also, turning off the cutoff at 20 kHz is generally not justified, as a person simply does not hear the higher frequencies.

There is a kind of “magic” EQ preset that can significantly enhance the sound.
This is not entirely true, in the first place, because each configuration taken separately (headphones, acoustics, sound card) has its own parameters (in particular, its amplitude-frequency characteristic). And therefore each configuration must have its own unique approach. Simply put, such an EQ preset exists, but it is different for different settings. Its essence lies in adjusting the frequency response of the path, that is, in “leveling out” unwanted voltage dips and surges.

Also, among people who are far from direct work with sound, it is very popular to set the graphic equalizer “with a tick”, which actually represents an increase in the level of the low and high frequency components, but at the same Time leads to muffled vocals and instruments, whose sound spectrum is in the mid-range region.

Before converting music to another format, you must “unzip” it to WAV
I would like to point out right away that WAV stands for PCM (pulse code modulation) data in a WAVE container (file with extension * .wav). This data is nothing more than a sequence of bits (zeros and ones) in groups of 16, 24 or 32 (depending on the bit depth), each of which is a binary code of the corresponding sample width (for For example, for 16 bits in decimal notation (these are values ​​from -32768 to +32768).

So the fact is that any sound processor, be it a filter or an encoder, generally works only with these values, that is, only with uncompressed data. This means that to convert audio from, say, FLAC to APE, you just need to decode FLAC to PCM first and then encode PCM to APE. It’s like repackaging files from ZIP to RAR, you need to unzip the ZIP first.

However, if you’re using a converter or just an advanced console encoder, intermediate to PCM conversion happens on the fly, sometimes even without writing to a temporary WAV file.

Myths of Digital Music Part 3

Myth 1

Different software players sound different (eg foobar2000 is better than Winamp, etc.)
To understand why this is not the case, you need to understand what a software player is. In fact, it is a decoder, drivers (optional), an output plugin (to one of the interfaces: ASIO, DirectSound, WASAPI. Etc.) and of course the GUI (graphical user interface). Since the decoder in 99.9% of cases works according to the standard algorithm, and the output plug-in is just a part of the program that transmits a stream to the sound card through one of the interfaces, the reason for differences can only be manipulative. But the thing is, the drivers are usually disabled by default (or should be disabled, since the main thing for a good player is to be able to transmit the sound in its “original” form). As a result, only possibilities can be compared here. processing and output, which, by the way, is often unnecessary. But even if there is such a need, this is already a comparison of handlers, not players.

Here I would also like to mention my article on how to configure sound output on a computer and perhaps annoy users who admire the “colossal” changes in sound after the configuration described in it; in 95% of cases, this is self-hypnosis (except, of course, when during setup some “enhancer” or other driver was turned off and messed up the whole picture). Unfortunately, the gains from all of these tweaks with ReplayGain, resamplers, and limiters are slim. Read more in the article “One more time about the sad truth: where does good sound really come from?” …

Myth 2

Different versions of drivers sound different

This statement is based on a banal ignorance of the principles of sound cards. A driver is software necessary for the effective interaction of a device with the operating system, and generally also provides a graphical user interface to control the device, its settings, etc. A sound card driver ensures that the sound card is recognized as a Windows sound device. , informs the operating system of supported formats, provides uncompressed PCM stream transfer (in most cases) to the card, and also gives access to settings. Also, in the case of software processing (by means of the CPU), the controller can contain multiple DSPs (controllers). So, with effects and processing turned off in the first place, if the driver doesn’t provide accurate PCM transfer to the card, this is considered a fatal error. critical error. And it happens extremely rare. On the other hand, the differences between the controllers can be in the updating of the processing algorithms (resamplers, effects), although this does not happen often either. Also, effects and any controller processing should be excluded for maximum quality.

Therefore, driver updates are mainly focused on improving stability and fixing handling errors. In our case, neither one nor the other affects the playback quality, so in 999 cases out of 1000 the driver does not affect the sound.

Myths of Digital Music Part 2

DVD-Audio sounds better than Audio CD (24-bit vs. 16, 96 kHz vs. 44.1, etc.)

Unfortunately, people generally only look at numbers and rarely think about the impact of a particular parameter on objective quality.

Let’s first consider the bit depth. This parameter is only responsible for the dynamic range, that is, the difference between the lowest and highest sounds (in dB). In digital audio, the maximum level is 0 dBFS (FS – full scale), and the minimum is limited by the noise level, that is, in fact, the dynamic range in absolute value is equal to the noise level. For 16-bit audio, the dynamic range is calculated as 20 × log 10 2 16, which is 96.33 wB. The dynamic range of a symphony orchestra is up to 75 dB (mainly around 40-50 dB).

Now let’s imagine the actual conditions. The noise level in the room is about 40 dB (do not forget that dB is a relative value. In this case, the hearing threshold is taken as 0 dB), the maximum volume of music reaches 110 dB (so that no discomfort) – we get a difference of 70 dB. So it turns out that a dynamic range of more than 70 dB in this case is simply useless. That is, at a higher range, loud sounds will reach the pain threshold or soft sounds will be absorbed by the surrounding noise. It is very difficult to achieve an ambient noise level of less than 15 dB (since the volume of human breath and other noises caused by human physiology is at this level), as a result, a range of 95 dB for listening music is completely sufficient.

But there is a “but” here. If you generate a clean tone with a frequency of, for example, 1 kHz and a level of -60 dBFS with a 16-bit quantization depth, and then you listen to it and compare it to the same signal, but generated in 24-bit format , you will hear the differences. The reason lies in the distortion of the waveform and the appearance of parasitic harmonics. But to eliminate this unpleasant effect, fortunately, there are Dithering and Noise Shaping technologies.

Now about the sample rate (sample rate, sample rate). This parameter is responsible for the time sampling rate and directly affects the maximum frequency of the signal that can be described by this audio representation. According to Kotelnikov’s theorem, it is equal to half the sampling frequency. That is, for a typical sampling frequency of 44100 Hz, the maximum frequency of the signal components is 22050 Hz. The maximum frequency. that is perceived by the human ear, just above 20,000 Hz (and even then, at birth; as we age, the threshold drops to 16,000 Hz).

Myths of digital music

Lossy codecs (MP3 and others) can cope with modern electronic music, but cannot efficiently encode classical (academic), live and instrumental music.

The “irony of fate” here is that everything is actually the exact opposite. As you know, academic music in the vast majority of cases follows melodic and harmonic principles, as well as instrumental composition. From a mathematical point of view, this leads to a relatively simple harmonic composition of the music. So the predominance of consonances produces fewer side harmonics: for example, for the fifth (the interval in which the fundamental frequencies of two sounds differ by one and a half times), each second harmonic will be common for two sounds, for a fourth, where the frequencies differ by one third, every third, etc. Furthermore, the presence of fixed frequency ratios, due to the use of equal temperament, also simplifies the spectral composition of classical music.

The factors listed above lead to the fact that classical music is much easier to compress, mainly in a purely mathematical way. If you remember, mathematical compression works by removing redundancy (describing similar pieces of information using fewer bits), as well as predicting (so-called predictors predict the behavior of the signal, and then only the deviation of the actual signal from the predicted one is encoded; the more exactly they match, fewer bits are needed for encoding). In this case, relatively simple spectral composition and harmonicity lead to high redundancy, the removal of which provides a significant degree of compression, and a small number of bursts and noise components (which are random and unpredictable signals) leads to good predictability. mathematics the vast majority of information. Not to mention the relatively low average loudness of classic tracks and the frequent gaps of silence, which require virtually no information to encode. As a result, we can compress without loss, for example,

So, first of all, the fact is that the mathematical compression underlying lossless encoding is also one of the stages of lossy encoding (read Understanding MP3 encoding). And secondly, since lossy uses the Fourier transform (decomposition of the signal into harmonics), the simplicity of the spectral composition even makes the encoder’s job twice as easy. As a result, when comparing the original and encoded sample of classical music in a blind test, we are surprised to find that we cannot find any difference, even at a relatively low bit rate. And the funny thing is that when we start to completely lower the encoding bit rate, the first thing that detects the difference is the background noise in the recording.

As for electronic music, encoders have a hard time: noise components have minimal redundancy and, along with jerky jumps (some sawtooth pulses), are extremely unpredictable signals (for encoders that are “sharp “by natural sounds that behave completely differently), the direct and inverse Fourier transform with the rejection of individual harmonics by the psychoacoustic model inevitably produces pre and post echo effects, the audibility of which is not always easy to evaluate for the encoder … Add to this a high level of HF Components, and you get a lot of killer samples that even the most advanced encoders can’t handle at medium-low bit rates – oddly enough, it’s somewhere between the electronic music.

Also amusing are the opinions of “experienced listeners” and musicians, who, with a complete misunderstanding of the principles of lossy encoding, begin to claim that they hear how the instruments in music, after encoding, begin to falsify, the frequencies float, etc. perhaps it would still be true for detonating antediluvian cassette players, but in digital audio everything is exact: the frequency component remains or is discarded, there is simply no need to change the key.

Also: a person’s ear for music does not at all mean that they have good frequency hearing (for example, the ability to perceive frequencies> 16 kHz, which decreases with age) and does not make it easier for them to search for encoding artifacts at a loss. Since distortion has a very specific character and requires the expertise of blindly comparing lossy audio, you need to know