maybe add recordings of my own saxophones to default soundfont?

• Feb 9, 2018 - 18:47

Last weekend lasconic and thomas were talking about soundfonts for 2.2 and I mentioned a while ago I experimented with making my own soundfont from recording myself playing each pitch on my tenor sax...take a listen to how it sounds in musescore on reunion: https://www.youtube.com/watch?v=mgKPAkUIXG8

If people like that, I would be happy in making such a soundfont for all my saxophones (bass, bari, tenor, alto, soprano, sopranino) using a sample from each pitch to be included in 2.2 if desired. If that is wanted, I would need to know the specs for recording (would that be 48k mono 16bit?) maybe other specifics like typically how long should the loop be, and maybe for consistency which program should i use for making the soundfont (I was using some sf2 editor earlier). Another thing to consider is I have a handful of different mouthpieces for tenor and bari...I could at least do a standard classical mouthpiece version and a wide-spacing version.


Comments

I'm no soundfont expert so I'll leave that to others to discuss the technical details, but I will say your soundfont sounds great, and it was fun to hear on Reunion! Sure hope we could get slurs happening someday, though.

Eric, your tenor sax example sounds very good indeed! I am developing the new default SoundFont for MuseScore and have been looking for freely-licensed samples to improve the instrument sounds. I have had a hard time finding good saxophone samples, so the work you have offered to do would be a most welcome addition to the upgraded SoundFont.

Here is what we would be looking for, in descending order of priority:
Required: Record notes across the entire range of the instrument at mezzo-forte to forte dynamic level (we want the notes bright but not forceful), ideally sampling every half-step. Having every note sampled will allow me to use sample shifting to create variants for each solo instrument. For example, a single set of alto sax samples will allow me to create three alto sax instruments that can all be layered together with no flanging effect on the unisons.

Required: Notes should be played with a quick attack, but they shouldn’t be accented. This is necessary to facilitate quick passages in notated scores. Notes that take too long to start often sound good in slower passages, but completely fail when faster notes must be played. “Slow” variants of solo instruments can be simulated using the envelope to create a longer attack, but “fast” variants cannot be simulated from slow samples.

Required: For sustained instruments, each note should be a steady tone that lasts for approx. 5 seconds. I can loop the samples to allow the notes to sustain indefinitely. Try to avoid any change in the volume or pitch (except for vibrato) during the course of the 5 seconds to provide for smoother loops. For saxophone samples, notes that start with no vibrato and then fall into a nice vibrato seem to work well for multiple music styles, but I'll let your musician's ear be the judge of the best sound to capture.

Required: Samples should be recorded at 16-bit, 44.1 KHz. Anything higher bitrate & bit depth will be downsampled to save memory. Samples can be sent as separate WAV files, and I can take care of further editing, such as looping, etc. I will accept SF2 files as well. Please label the samples according to the instrument, dynamic and note, e.g. “Alto Sax mF C5.wav”. You can either post download links for your completed samples here or send them to https://musescore.org/en/user/62809/contact.

Ideal: Samples should be recorded with a single microphone (mono) and minimum room reverberation if possible. Existing stereo samples will be accepted, but we will likely convert them to mono for inclusion in the new SoundFont.

Optional: If you have time to burn, and want to create an even more realistic instrument, you can record samples at other dynamic levels as well. Fortissimo and piano samples would be the most useful to have in addition to the required mF-F recordings. In cases where only the loud samples are provided, I will attempt to simulate quieter samples by using a lowpass filter to give the instrument a change of tone at lower dynamic levels.

Please let me know if you have any questions or if I can be of assistance in any way. We are looking to license all new samples as CC0, but please let me know if this would be an issue for you. All samples and their authors will be credited per-preset in the included documentation for the new SoundFont.

Thanks again for offering to contribute your talents and resources to help make MuseScore better!

In reply to by s.chriscollins

Thanks s.chriscollins for those detailed specs. Regarding vibrato, I think it is better to have the sample without vibrato, because vibrato can always be easily similated, but not the other way around (trying to remove vibrato from a vibrato sample).

Yes CC0 is fine.

I guess I will have the mic about a meter diagonally above my bell.

I was unaware that MuseScore 2.2 release date was today. All I remember being told was that 2.2 would be released before summer. I suppose this is what I get for not paying attention to other forums and being on irc all the time. And I got distracted by students asking questions about GSoC ideas.

What I'll do instead is just make my own soundfont and whoever wants it can put it at the top of the synthesizer window in musescore so it will be used for saxophone channels. But I want to know more info about how the editing should be done. For one, are mf and f samples normalized in any manner? And how do you normalize samples between instruments? Do you normalize according for equal loudness between instruments or just normalize to peak? Also is there a particular editor you use for making the sf3. I remember for sf2 I used an editor, but I don't know of an editor for sf3.

In reply to by ericfontainejazz

I wasn't able to create a lot of new instruments in time for 2.2 due to licensing issues with my existing ready-to-use samples, which caused me to have to do a ton of sample editing/looping. I've got some really nice strings on the way, and we have some new sample sources that have recently become available to us that we will be using for further development of the SoundFont.

You don't have to worry about creating the SoundFont. Just get me the samples, and I will make it happen. If you want to normalize the samples, that would be helpful as well. If you do so, it would also be useful to have a list of how much each sample was amplified if it's not too much trouble.

Thanks again for creating these saxophone samples!

In reply to by s.chriscollins

I recorded tenor with my Selmer C* mouthpiece, which is a standard non-jazz ensemble mouthpiece, but after going through it I regret not doing it on one of the mouthpieces I play more commonly. After listening carefully back I think I might redo many of these. Looking at pitch tracker on Sonic Analyzer, I see most notes drift around, but I pick the best I could get at the momement. Also I don't think I always got a identical attack, particularly on mF. But take a listen and let me know anything in particular before I try it again or do other saxophones:

https://drive.google.com/file/d/1U5N_EcH9To5CSfxIl5LUXxEgFJ4xrAwV/view?…

A lot I could do better. I'll leave my mic setup exactly where it is, so I can redo the bad ones. Oh and anything higher than C5 and altissimo starts to get out of tune, so I'll have to redo those for sure.

In reply to by s.chriscollins

I did Alto Sax forte & mezzo:

https://drive.google.com/file/d/1kiXGAQTTxrui-6m8ZZ1QXPIaz1jKVO1b/view?…

i'm much more confident with this than the tenor one. I made sure to record directly into sonic visualizer and connect myself to a more accurate tuning program. Please take a look at that.

Btw, I notice my mezzo takes generally have a slower attack than the forte ones, even though I was trying not to. I wonder if it makes sense to edit them such that the full sound happens at a fixed precise number of milliseconds after initiation.

I'll get around to the other saxophones this week. But let me know how that alto is.

In reply to by s.chriscollins

Also, the wav samples from the tenor I recorded on my most commonly-used mouthpiece (P.Mauriat mouthpiece) from 2015 and don't sound too bad, so maybe download them too and take a listen and see if can be salvaged: https://drive.google.com/open?id=1lXBSFuXIH6XGtieTNY3RUFZu-BfP6g-n

And the .sf2 file I made from those is https://drive.google.com/open?id=16EHpaNJTfD-rHmh_Po4DAPTeCLCa26zO

Listening back...they seem to be in the mF, so maybe they can still be used I could record Forte version on that mouthpiece too.

In reply to by s.chriscollins

I did Bari and feel fairly confident about it, both the attacks and sustained pitch intonation: https://drive.google.com/open?id=1tJTk2nbkssg9zmh2WvBFX2m7zg28SHwG

Another thing...I assume you will be adjusting loudness levels to be consistent? Unfortunately i can't really keep a constant idea of what Forte is or what mF is in my head while recording...listening back I hear quite a bit of variation.

In reply to by ericfontainejazz

Although Sonic Visualiser's recording capabilities are limited, with a little trick you can get it to display simultaneously in real-time a pitch spectrogram and a db waveform which can help you stay more consistently within a targeted dynamic range (e.g., mezzo-forte to forte range).

Start recording then switch to the "Waveform" tab on the right of the GUI, change Color to "white", so it stands out, Scale to "db", and Source to "mean". To make the spectrogram stand out more as well, select "Col" in the second Scale drop-down menu in the Spectrogram tab. Make sure the last tab your select is "Waveform" otherwise you won't see the db/meter being graphed in real-time.

The image below is me singing two notes, recorded in Sonic Visualiser. In real-time, I can watch the recording level, as the white line, and the fundamental pitch of my voice, as the sunburst-theme spectrogram.

spectogram-and-db-meter.png

In reply to by ericfontainejazz

Your welcome. I often use Sonic Visualiser for transcribing music. I haven't really used the recording feature until now. I will contact the developer to suggest that he add the capability of setting up the tabs before recording starts. Currently, the settings I suggested above, which can be done within a few seconds, has to be done will the "reel is running", the program is recording.

In reply to by Sambaji

Here is a Sonic Visualiser template of my suggested settings. Download and unzip the file to C:\Users\Your name\AppData\Roaming\sonic-visualiser\Sonic Visualiser\templates

I had to zip the file so that Musescore.org will accept it for upload.

Under "File > Preferences > Session Template" set the default template to "recording", the name of my template. Restart the application. Now when you hit "record" the template will automatically load. You just have to switch to the Waveform tab so that recording level is visible. Alternatively, you create your own template based on your settings, under File > "export session as template".

Attachment Size
recording.zip 822 bytes

In reply to by Sambaji

Here is dual-pane template. The spectrogram is in the upper frame, the db/meter is in the bottom. You may want to drag the bottom of the upper frame down so that more of the spectrogram is showing. If you are recording lower notes, try changing the second Bin drop-menu on the spectrogram tab in the upper frame from Linear to Log. Log will make the lower notes more apparent. Also changing the first Bin drop-menu to Frequency may give you a more sensitive reading of the pitch. You can also change the spectrogram's frequency range by right clicking on the lower right vertical bar of its display. Ideally, the range should be close to the lowest and highest note your are recording so you can easily see any pitch changes. If you make changes to the settings you want to keep, simply" export the session as template" and then select to make it the default.

Attachment Size
recordingdualpane.zip 935 bytes

In reply to by Sambaji

thanks a bunch...using this dual pane setup now and it is just what I need. Although one slight weakness of sonic visualizer is that the rendering of the spectrum & dB chart isn't quite instantaneous, but rather seems to render one-second chunks at a time. I'm guessing this is by design... Unless there is some magical parameter that I can lower that...if not it would be great if sonic visualizer could be made to have that.

In reply to by ericfontainejazz

Maybe try out some free audio analyzers such as Friture ( http://friture.org/ ) that are designed specifically for real-time feedback. Friture doesn't have recording capabilities so you would have to run it with another recording program such as Audacity (https://www.audacityteam.org/). I would set the 2d spectrogram's FFT size to 8192 for higher res. Click on the wrench and hammer icon next to each drop-down menu to access each pane's/"dock's" settings. If you click on Add Dock in the upper left of the Gui, it will automatically add a Scope, helpful for seeing recording levels. Or you could just change the type of graph shown in a panel by selecting another option in the drop-down menu. I prefer to have just three docks so that each are larger. The db levels are shown on the left side regardless.

In reply to by Sambaji

Friture looks great. ... Trying to install now although arch doesn't seem to have a package for it and pip is missing dependencies somehow and not automatically installing them...hopefully I will get it working. But I don't see an option to view frequency in log matching up to a grid corresponding to musical pitches. That is what I much prefer.

Anyway lingot tuner plus sonic visualizer does what I need eve

Do you still have an unanswered question? Please log in first to post your question.