More on legato

Posted 8 years ago

Hi,
last couple of days I was investigating in legato realization since Isaac Weiss was not happy with the current sound and I also thought that there might be some improvement. Even though I didn't found it as problematic as him. But here are my results on that topic. This post uses a lot of spectrograms - if you're unfamiliar with this type of diagrams you might have a look here.

A spectrogram basically displays frequency on the y-axis and time on the x-axis. While the intensity of each frequency is given by a color spectrum.

Everything I uploaded before contained the default reverb of MuseScore. This makes hearing the differences more clearly more difficult (since everything gets "washed"/"blurred" together).

First I started of my experimentation by actually using an array of gain values to make sure I fading is nice and smooth and 100% under my control. I wanted to try out two different types of fades - a linear one and an equal power one. Equal power means that you don't have a value of 0.5 in the middle but a value of 0.707 (-3dB). If the signals are completely uncorrelated this will avoid a dip in the signal. But actually this doesn't make so much sense since the same instrument is in a lot of times highly correlated to itself - but I had to try it out! ;)

But lets start with some sound. In the name you can see which instrument it is, the number tells you the fade sample count, than it says if it a linear fade or a equal power one and last but not least if it uses reverb or not. It always first played with the legato feature enabled and after that without it.

First without reverb:
Flute_1200_eqp_nRev.mp3
Flute_1200_lin_nRev.mp3
Flute_3000_eqp_nRev.mp3
Flute_3000_lin_nRev.mp3
Violin_1200_eqp_nRev.mp3
Violin_1200_lin_nRev.mp3
Violin_3000_eqp_nRev.mp3
Violin_3000_lin_nRev.mp3

And for you interest with reverb:
Flute_1200_eqp_Rev.mp3
Flute_1200_lin_Rev.mp3
Flute_3000_eqp_Rev.mp3
Flute_3000_lin_Rev.mp3
Violin_1200_eqp_Rev.mp3
Violin_1200_lin_Rev.mp3
Violin_3000_eqp_Rev.mp3
Violin_3000_lin_Rev.mp3

As you can hear linear sounds better in general than the equal power one. That is especially because the sound of violin and flute is strongly correlated to itself. 3000 Samples seems a better number than the 1200 Samples I started of. I choose 3000 Samples because that was about how long it takes in the violin samples to build up the sound. I still don't like that the second note in a legato sticks out. That is because the sample of most sounds is a bit louder during the loop than in the beginning. But that is something I think we should fix in the soundfont.

But I wanted to have a look how Native Instruments does that in their Session Strings product. So lets have a look:
NI_SessionStrings_Legato.jpg
As you can see the release/fade out takes a much longer time than the attack/fade in. Because I looked at the first overtone I also thought it would start to fade out after the fade in finished.

This is what it would sound like:
Flute_mimic_NI_nRev.mp3
Violin_mimic_NI_nRev.mp3

With reverb:
Flute_mimic_NI_Rev.mp3
Violin_mimic_NI_Rev.mp3

But now you clearly hear two tones - that is not good. I took a deeper look with another spectrogram setting:
NI_SessionStrings_Legato_no_reassign1024points.jpg

Hear you can clearly see that it starts to fade out as soon as the new note appears. This is what it sounds like:
Flute_mimic_NI2_EQP_nRev.mp3
Flute_mimic_NI2_nRev.mp3
Violin_mimic_NI2_EQP_nRev.mp3
Violin_mimic_NI2_nRev.mp3

Flute_mimic_NI2_EQP_Rev.mp3
Flute_mimic_NI2_Rev.mp3
Violin_mimic_NI2_EQP_Rev.mp3
Violin_mimic_NI2_Rev.mp3

I also added equal power fades here - just to have look how it would sound. Thins the time is higher it doesn't sound so bad.

I also had a look if what I created mimics NI in the right way:
My first try with fade out starting after fade in:
Violin_1stTryReimplement_NI.jpg
2nd attempt with fade out starting at the same time as fade in
Violin_2ndTryReimplement_NI.jpg
2nd attempt with equal power fades:
Violin_2ndTryReimplement_EQP_NI.jpg

I also had a look what legato looks like as played on real instruments:
Legato on a cello Cello.jpg
Legato on a flute FluteLegato.jpg

I got the samples for this website and this one.
Interestingly you can see a bend for both instruments during legato playback. I thought that this would just happen for voice or is a different technique for bowed instruments. But that is something I'll work on next to realize bending into the next note.

But I think 3000 Samples and linear fade sounds good to me. The NI approach is also not band in my opinion. But I'm interested what you think sounds best!


Comments

Thanks so much for the deeper investigation! I agree that 3000 samples and linear fade sounds the best of these, especially with the reverb. I think there must be something unique about the samples for the strings in FluidR3Mono that makes them super legato at the beginning of the attack and less so at the loop point, but I can live with this approach—especially given how dramatically better the results are with many of the other instruments (that baritone sax...). I'd be interested to hear how this sounds with the violin from, say, GeneralUser, which has a more conventional sharp attack at the beginning of the samples. I suspect it would come out very well.