Music: Practice & Theory Stack Exchange is a question and answer site for musicians, students, and enthusiasts. It only takes a minute to sign up.
Sign up to join this community
I want to change the pitch of an audio track and try to keep the highest quality possible, so when I click on processor algorithm setup, I have few options :
Standard method (time stretches and pitch shifts without the aid of beat markers) Smooth method HQ Universal ect ect... can someone explain to me what beat markers mean in pitch shifting ? and what would be the best quality ? is it the standard or the HQ Universal ?
can someone explain to me what beat markers mean in pitch shifting ?
Yes! I recommend reading the User's Manual.
From the user's point of view, there are basically two types of time-stretching algorithms: "sliced" and "elastic":
(1) Slice-and-move/repitch, like the Propellerheads ReCycle product with its REX file format. Best suited for individual isolated rhythmic instrument tracks: drum and percussion loops, guitar chord strumming loops, bassline loops, riffs, etc. Not really suitable for full mixes and vocals.
In the good old Propellerheads ReCycle system, the software detects note/hit attack transients and places beat markers on them (and you can adjust the detection sensitivity and IIRC you can also manually place markers), and then the audio is basically chopped up at the beat markers, creating audio "slices". Time-stretching is done by moving the slices on the timeline, and pitch shifting is done by playing the sound inside each slice slower or faster. If played at exactly the original timing and pitch, you wouldn't notice any differences, but when time-stretching, the slices are spread around, keeping the beat markers on the sequencer's beats (or other desired time locations), as if someone had played the slices as individual samples on a sampler. Playing the slices at a higher tempo is easier, but when you slow down the tempo, you move the slices further apart, and then you get silent gaps between the slices. For some drum tracks this might not be a problem, if every note slice fades out to silence anyway. But for more continuous tracks, what to do? One method is to fill the gaps by ping-pong looping each slice backwards/forwards until the next slice.
Slicing does not work very well for full mixes, slow pad-like sounds, or vocals, because the slice-and-move or slice-and-repitch method (with the gaps and all) usually makes the audio sound too choppy and vocals unnaturally hiccupy.
(2) "Elastic" algorithms (which themselves are further subdivided to time-domain overlap-add and frequency-domain spectral methods, but from the user's point of view they can be treated somewhat similarly). These algorithms are more suitable for full mixes and vocals, because they don't suffer from the hiccup problem. But for sharp attacks like drums, they might ruin the punchiness. You don't want to un-sharpen the attacks of your drums. The elastic algorithms also need more processing power than slicing, so they'll be heavier on your computer (CPU).
Elastic methods treat audio like an elastic rubber-band, smearing everything more or less equally, not making sharp cuts or slices anywhere but trying to smoothly slide from one spectrum (timbre) to another. Some algorithms might try to somehow automagically "understand" what's happening with the audio, and which spots are important to keep less smeared. Elastic algorithms are better for full mixes and vocals.
Beat markers are important for these algorithms as well, in order to preserve the proper rhythmic feel. Or to do things like audio quantization and "groove quantization".
From pitch shifting point of view, beat markers are the spots where the virtual keyboard player plays a new note on the sampler, if you imagine that it's like having chopped up the audio to individual samples in a sampler.
and what would be the best quality ? is it the standard or the HQ Universal ?
It really depends on many things, and you have to try it out. What works for some type of material might not work for another. There's no "universal" setting for everything.
The reason different algorithms exist is because there is no "best" - they all are good at different things. And the specific algorithms those names refer to may vary across products.
For the genre of music you want to create, your best bet is to try them all and see which works best. Does it cope with your beats, your highs and lows, rapid changes, etc.?
There are two general methods in digital signal processing for time/pitch modification: time domain and frequency domain.
Time domain methods generally stretch sustained note sound by whole pitch periods (add one or remove one time slice of 1.0/pitch_frequency in duration). This can throw off the precise timing of notes, and may remove few-millisecond-sized (but interesting) sound features due to deletion.
Frequency domain methods generally stretch everything equally in time, which does not throw off precise timing as much, but potentially can smear the timbre of note transients (attach and percussion) and distort voice formant.
So, it's a trade-off. Choose your preferred artifact.
