The quality of transposing samples in Ableton Live 8 tools and free plugins
Did you notice, after adjusting the pitch of the sample loaded in the sampler or drum machine, that dirt and harmonic harmonics were added? Such a degradation of sound occurs when the sampling frequency changes, which inevitably occurs when transposing or warping an audio file, and is associated with the aliasing effect — the reflection of the spectral components from the Nyquist frequency. To avoid this, interpolation and filtering algorithms are used.
In this article, using the elementary test, we compared such algorithms in the tools Impulse, Simpler, Sampler and the Ableton Live 8 sequencer itself, as well as with some free popular Vst plugins.
Experimental technique
')
The experiment was conducted on a laptop Lenovo R500: Intel Core 2 Duo T8570 processor, 2 Gb of RAM. The sampling rate, in the settings of Ableton Live, the test sample, and the exported file is 44100 Hz. The test file is a sinusoidal signal at a frequency of 15 kHz, 16 bits wide, which is loaded into the sampler, is tied to the note C3, and the playing area is expanded to the entire keyboard range. Each sampler under test loses the same MIDI sequence — the C major gamma is two octaves up, and then two down from the C3 note. In the case of loading a sample to an audio track of a sequencer or to Impulse, using the automation of the tune parameter, the equivalent sequence is played with the timing of the sample start and the corresponding midi note of the sequence. The output volumes of the tested tracks are set to an accuracy of ± 0.1 dB, with such a margin that there is no overload on the master channel. Sound files that are the result of playing a MIDI sequence or each instrument equivalent to it in solo mode are exported with a bit depth of 24 bits and loaded into Adobe Audition. In the “spectral frequency display” mode, spectrograms are compared that demonstrate changes in the sound spectrum over time. The indications of the CPU utilization indicator in Ableton Live with the operation of a tool are given in the results.
Results and its discussion
For comparison, the ideal sound is taken as a reference, and accordingly the export spectrogram from the DiscoDSP HighLife sampler in mastering mode. A black background corresponds to ideality (no noise), at each moment of time only one frequency can be seen and heard, except for frequencies exceeding half the sampling rate - they should not be. Unfortunately, this ideality has its drawbacks, as it is intended only for export. I tried to play a test sequence of notes in real time. The CPU utilization indicator is up to 380%. I did not think that it happens.
Now imagine that we loaded the drums samples into Impulse or Drum Rack, which Simpler uses, and tried and changed their pitch a bit. The result is obvious, judging by the spectrograms, and by ear, there will be noticeable distortion. Impulse is a bit worse, as it smears the spectrum of the sample a little. Apparently this is why Ableton developers recommend using Drum Rack instead of Impulse.
Well, we'll use Drum Rack, and it seems that every Simpler in his cell can be converted to a Sampler. In this case, it will be possible to set the interpolation algorithm. By default, it takes the value normal, which, as the test shows, corresponds to Simpler. Switching to the Good mode gave improvements, but to the ear, the distortions are still noticeable. Best algorithm has pleased almost inaudible distortions, on a complex signal, they will be imperceptible than in comparison with a pure sinusoid. The absence of an interpolation algorithm can be an effective creative tool for creating glitches and dissonant dirty digital textures, but not for high-quality reproduction of a transposed sample.
The audio sequence interpolation algorithm of the sequencer in the HQ On mode turned out to be interesting, it is more favorable by hearing than the Best Samplera algorithm, and as you can see on the spectrogram, it is noticeably different: weaker tones reflected from the Nyquist frequency, but also higher disguises, but adds total dirt.
It should be noted that the two nuclear laptop of six years ago did not notice an increase in the load on the processor, and in all cases the indicator did not exceed the 7% barrier, which confirms the positioning of Live as a tool for live performance. The second highly personal conclusion is that it is better to use the audio track directly for downloading samples, rather than any native program tool. Of course, if there is no goal to noticeably change its sound with the help of ADSR amplitude, pitch, filter, frequency modulation (present in Sampler), etc ... But you can achieve a cleaner sound using third-party instruments, which we are going to review.
Probably the first most popular on the network is the Cakewalk Sfz + plugin, which is positioned for downloading from SoundFont banks. It has long been distributed free of charge (the version number of the program is 1.0). Apparently, the “greats” knew a lot about aliasing, in the era of quite primitive computers from our point of view. Therefore, Sfz + has nine levels of complexity of the algorithm for working with the sample, and the weakest processor will cope with the simplest. Nowadays, you can safely use the option "quality: 72", which corresponds to the highest quality playback, the result of its action is shown in the figure. He is a little less than the ideal (DiscoDSP HighLife in mastering mode), but everything is excellent by ear. The disadvantage of Sfz + is the presence of only basic tools for working with the sample, among which there is no possibility of setting the place of its start, loop points and playback options. There are also problems with using a plugin on 64 bit OS.
DiscoDSP HighLife - an object of imitation in our tests. The second interpolation algorithm “bounce mode” is not much inferior in quality, but it is also difficult to use it in real time due to the high processor load. The “realtime mode” algorithm, judging by the spectrograms, coincides with the “good” algorithm in Sampler. The functionality of the plugin is somewhat wider than that of Sfz +.
Next applicant: Vember Audio Shorcircuit 1. Distortion during transposition is not audible and, judging by the spectrogram, its algorithm takes place somewhere between the ideal and the Good Sampler algorithm. The plugin has a huge set of excellent quality filters that even Native Instrument’s Kontakt cannot boast of. The interface is very simple and easy to learn. Almost a barrel of honey, if not a spoon of tar, which can be read on the user forum . This plugin delays and eats the attack of the sample. Moreover, in version 1.0.15 this was not (so if you know where to get this particular version and are ready to share, I would be very grateful to you). But it is worth noting that the delay of about 32 samples is not critical. In Live this can be compensated. Oiled attack is not correct, but at the hearing it is not noticeable. The next version of Shorcircuit 2, pleased with the improved graphical interface. Tests have shown that there is no smear attack, but the delay is still in place. It is a pity that the developers have left this beautiful thing unfinished. By the way, if I'm not mistaken, one of the developers of this sampler was in the Ableton team, and now he is engaged in a project in a new collaboration with a similar ideological approach, but with some additional and very convenient features: bitwig.com .
Conclusion
I hope that this article will help in creating better sounding phonograms and in no case will not become an object for the emergence of disputes and the next holivar on the topic of plug-ins and music software.
If you want more comparisons with other samplers, then at the end there will be links to the methodology and results that are not reflected in this article. It is inexpedient to do tests of all existing plug-ins, and some are not intentionally included in the article (for example, Kontakt and Battery), because I want to encourage the reader to conduct an experiment himself. To do this, simply load the 15 kilohertz sine or any high-frequency audio file into your favorite sampler and run the keys two octaves up. Your hearing will not fail.
Those who carefully examined the spectrograms, noticed that when transposing the sample per octave, leads to less distortion, because the interpolation algorithm is simplified due to the change in the sampling rate of the sample in a multiple of two the number of times. If the sampling frequency of the sample and the program settings do not match, this effect will disappear, and the overall test results will be worse. The solution could be to use codecs for offline conversion of the audio file sample rate even before using them in a project. There are many tools to do this, and one of the best is the free plugin from Voxengo r8brain free .
I want to call for the study of the mat part, at least at the level at which it is available to you, those who have not done it yet (this is common among digital composers). It is worth starting with the Nyquist theorem and reading the software manual, or other documents, such as Ableton's “Audio Engine Fact Sheet”, which describe operations detrimental to sound. Having the appropriate knowledge will allow you to avoid unwanted sound degradation when creating compositions.
Clean you sound!
Links
Methods and results of sampler tests:
www.discodsp.com/highlife/aliasing
jeskola.net/xs1/content/test
www.maz-sound.de/resampling
www.simonv.com/tutorials/quality.php
Methods and test results of algorithms for converting the sampling frequency of programs for working with audio:
src.infinitewave.ca
Mat part:
www.digital-recordings.com/publ/pubneq.html - Nyquist theorem in the context of sound application
en.wikipedia.org/wiki/Aliasing - aliasing
kunz.corrupt.ch/dsp - something else useful in digital synthesis and creating effects for the "savvy"
In this article, using the elementary test, we compared such algorithms in the tools Impulse, Simpler, Sampler and the Ableton Live 8 sequencer itself, as well as with some free popular Vst plugins.
Experimental technique
')
The experiment was conducted on a laptop Lenovo R500: Intel Core 2 Duo T8570 processor, 2 Gb of RAM. The sampling rate, in the settings of Ableton Live, the test sample, and the exported file is 44100 Hz. The test file is a sinusoidal signal at a frequency of 15 kHz, 16 bits wide, which is loaded into the sampler, is tied to the note C3, and the playing area is expanded to the entire keyboard range. Each sampler under test loses the same MIDI sequence — the C major gamma is two octaves up, and then two down from the C3 note. In the case of loading a sample to an audio track of a sequencer or to Impulse, using the automation of the tune parameter, the equivalent sequence is played with the timing of the sample start and the corresponding midi note of the sequence. The output volumes of the tested tracks are set to an accuracy of ± 0.1 dB, with such a margin that there is no overload on the master channel. Sound files that are the result of playing a MIDI sequence or each instrument equivalent to it in solo mode are exported with a bit depth of 24 bits and loaded into Adobe Audition. In the “spectral frequency display” mode, spectrograms are compared that demonstrate changes in the sound spectrum over time. The indications of the CPU utilization indicator in Ableton Live with the operation of a tool are given in the results.
Results and its discussion
| Ableton Live 8 Tools | Free Vst Samplers |
Sampler: best CPU algorithm <7% Sampler: good CPU algorithm <7%  Sampler: normal CPU algorithm <7% (just like in Simpler and Sequencer, the HQ algorithm is turned off)  Sampler: no interpolation CPU <7% algorithm  Impulse CPU <7%  Sequencer: HQ algorithm enabled CPU <7%  | DiscoDSP: mastering CPU> 350% algorithm DiscoDSP: bounce CPU algorithm> 50%  DiscoDSP: realtime CPU algorithm <7%  Cakewalk (rgc audio) Sfz +: Algorithm 72 CPU <7%  Vember audio Shortcircuit 1 CPU <7%  |
For comparison, the ideal sound is taken as a reference, and accordingly the export spectrogram from the DiscoDSP HighLife sampler in mastering mode. A black background corresponds to ideality (no noise), at each moment of time only one frequency can be seen and heard, except for frequencies exceeding half the sampling rate - they should not be. Unfortunately, this ideality has its drawbacks, as it is intended only for export. I tried to play a test sequence of notes in real time. The CPU utilization indicator is up to 380%. I did not think that it happens.
Now imagine that we loaded the drums samples into Impulse or Drum Rack, which Simpler uses, and tried and changed their pitch a bit. The result is obvious, judging by the spectrograms, and by ear, there will be noticeable distortion. Impulse is a bit worse, as it smears the spectrum of the sample a little. Apparently this is why Ableton developers recommend using Drum Rack instead of Impulse.
Well, we'll use Drum Rack, and it seems that every Simpler in his cell can be converted to a Sampler. In this case, it will be possible to set the interpolation algorithm. By default, it takes the value normal, which, as the test shows, corresponds to Simpler. Switching to the Good mode gave improvements, but to the ear, the distortions are still noticeable. Best algorithm has pleased almost inaudible distortions, on a complex signal, they will be imperceptible than in comparison with a pure sinusoid. The absence of an interpolation algorithm can be an effective creative tool for creating glitches and dissonant dirty digital textures, but not for high-quality reproduction of a transposed sample.
The audio sequence interpolation algorithm of the sequencer in the HQ On mode turned out to be interesting, it is more favorable by hearing than the Best Samplera algorithm, and as you can see on the spectrogram, it is noticeably different: weaker tones reflected from the Nyquist frequency, but also higher disguises, but adds total dirt.
It should be noted that the two nuclear laptop of six years ago did not notice an increase in the load on the processor, and in all cases the indicator did not exceed the 7% barrier, which confirms the positioning of Live as a tool for live performance. The second highly personal conclusion is that it is better to use the audio track directly for downloading samples, rather than any native program tool. Of course, if there is no goal to noticeably change its sound with the help of ADSR amplitude, pitch, filter, frequency modulation (present in Sampler), etc ... But you can achieve a cleaner sound using third-party instruments, which we are going to review.
Probably the first most popular on the network is the Cakewalk Sfz + plugin, which is positioned for downloading from SoundFont banks. It has long been distributed free of charge (the version number of the program is 1.0). Apparently, the “greats” knew a lot about aliasing, in the era of quite primitive computers from our point of view. Therefore, Sfz + has nine levels of complexity of the algorithm for working with the sample, and the weakest processor will cope with the simplest. Nowadays, you can safely use the option "quality: 72", which corresponds to the highest quality playback, the result of its action is shown in the figure. He is a little less than the ideal (DiscoDSP HighLife in mastering mode), but everything is excellent by ear. The disadvantage of Sfz + is the presence of only basic tools for working with the sample, among which there is no possibility of setting the place of its start, loop points and playback options. There are also problems with using a plugin on 64 bit OS.
DiscoDSP HighLife - an object of imitation in our tests. The second interpolation algorithm “bounce mode” is not much inferior in quality, but it is also difficult to use it in real time due to the high processor load. The “realtime mode” algorithm, judging by the spectrograms, coincides with the “good” algorithm in Sampler. The functionality of the plugin is somewhat wider than that of Sfz +.
Next applicant: Vember Audio Shorcircuit 1. Distortion during transposition is not audible and, judging by the spectrogram, its algorithm takes place somewhere between the ideal and the Good Sampler algorithm. The plugin has a huge set of excellent quality filters that even Native Instrument’s Kontakt cannot boast of. The interface is very simple and easy to learn. Almost a barrel of honey, if not a spoon of tar, which can be read on the user forum . This plugin delays and eats the attack of the sample. Moreover, in version 1.0.15 this was not (so if you know where to get this particular version and are ready to share, I would be very grateful to you). But it is worth noting that the delay of about 32 samples is not critical. In Live this can be compensated. Oiled attack is not correct, but at the hearing it is not noticeable. The next version of Shorcircuit 2, pleased with the improved graphical interface. Tests have shown that there is no smear attack, but the delay is still in place. It is a pity that the developers have left this beautiful thing unfinished. By the way, if I'm not mistaken, one of the developers of this sampler was in the Ableton team, and now he is engaged in a project in a new collaboration with a similar ideological approach, but with some additional and very convenient features: bitwig.com .
Conclusion
I hope that this article will help in creating better sounding phonograms and in no case will not become an object for the emergence of disputes and the next holivar on the topic of plug-ins and music software.
If you want more comparisons with other samplers, then at the end there will be links to the methodology and results that are not reflected in this article. It is inexpedient to do tests of all existing plug-ins, and some are not intentionally included in the article (for example, Kontakt and Battery), because I want to encourage the reader to conduct an experiment himself. To do this, simply load the 15 kilohertz sine or any high-frequency audio file into your favorite sampler and run the keys two octaves up. Your hearing will not fail.
Those who carefully examined the spectrograms, noticed that when transposing the sample per octave, leads to less distortion, because the interpolation algorithm is simplified due to the change in the sampling rate of the sample in a multiple of two the number of times. If the sampling frequency of the sample and the program settings do not match, this effect will disappear, and the overall test results will be worse. The solution could be to use codecs for offline conversion of the audio file sample rate even before using them in a project. There are many tools to do this, and one of the best is the free plugin from Voxengo r8brain free .
I want to call for the study of the mat part, at least at the level at which it is available to you, those who have not done it yet (this is common among digital composers). It is worth starting with the Nyquist theorem and reading the software manual, or other documents, such as Ableton's “Audio Engine Fact Sheet”, which describe operations detrimental to sound. Having the appropriate knowledge will allow you to avoid unwanted sound degradation when creating compositions.
Clean you sound!
Links
Methods and results of sampler tests:
www.discodsp.com/highlife/aliasing
jeskola.net/xs1/content/test
www.maz-sound.de/resampling
www.simonv.com/tutorials/quality.php
Methods and test results of algorithms for converting the sampling frequency of programs for working with audio:
src.infinitewave.ca
Mat part:
www.digital-recordings.com/publ/pubneq.html - Nyquist theorem in the context of sound application
en.wikipedia.org/wiki/Aliasing - aliasing
kunz.corrupt.ch/dsp - something else useful in digital synthesis and creating effects for the "savvy"
Source: https://habr.com/ru/post/219437/
All Articles