I'm not an expert in terms of micing and whatever, but I would in *no* way normalize anything at first.
There's really no sense in it because all you'll do by normalizing is raising levels, which is as well possible in any sampler there is - on a non-desctructive level.
Plus, especially when you're using many velocity zones (let's say something above 4), lining the velocity layers up is way easier than with normalized samples, when each and every velocity response would have to be recreated by your samplers velocity to amp level modulations.
You may however think about raising your input gain for the lower velocities. I'm not sure whether this is the "en vogue" method these days, I remember having a discussion with learjeff once, but I don't remember what the final verdict was (if any). I guess it depends on the used preamps as well. If they do a great job on amplifying, it might make sense to raise their gains on lower input levels, so you're probably not plagued as much by your recording environment's noise floor when playing quieter passages with the final sample set.
Of course, in this case the dynamic response has to be recreated inside the sampler as well, but in this case the lowered noise floor might show some advantages (others than when simply normalizing levels).
Anyways, how many velocity layers are you planning to do?
And how many keys are you going to sample?
IMO, if the piano itself sounds rather even all throughout, it'd be sufficient to sample seconds (or white keys only, whatever).
As far as looping goes, I'd go for an unlooped set first (way bigger but easier to handle) first and probably see if one can use less samples for a reduced, looped "ECO" version (I'd happily give the looping job a try, if you need someone to assist).
Personally, I'd love to see some unusual samples included, such as the pedal release "thunk", some release samples (they don't need to be meticulously sampled...). These things, when thrown in carefully, can be used for additional realism or strange sounds.
how to sample upright piano?
-
- KVRAF
- 13444 posts since 14 Nov, 2000 from Hannover / Germany
There are 3 kinds of people:
Those who can do maths and those who can't.
Those who can do maths and those who can't.
-
- KVRist
- 261 posts since 19 Jan, 2005
I have an unusual normalization method. I don't normalize everything to 0dBFS - I normalize the loudest sample and scale all the rest according to their RMS power (I may do a weighted combination of peak and RMS for the final version if it works better...). This, if pulled off correctly, allows mapping to the sampler with NO gain control whatsoever, besides velocity -> amp volume across all layers.
I have an advantage in that I can use MATLAB for this. I would not recommend doing it manually.
EDIT: Just noticed I may have given the wrong impression - although the method implied above does produce a usable instrument, it's not what I meant to say. Actually, once I've normalized the loudest sample, I don't scale the rest of the top velocity layer by RMS; I just multiply by the gain used to normalize the loudest strike, thus keeping the actual level differences. Then the RMS matching is used on the quieter velocity layers to match them to their respective ff samples. This method assumes that the top layer was recorded with uniform striking strength, and that the velocity-to-volume relationship is similar for all notes.
I have an advantage in that I can use MATLAB for this. I would not recommend doing it manually.
EDIT: Just noticed I may have given the wrong impression - although the method implied above does produce a usable instrument, it's not what I meant to say. Actually, once I've normalized the loudest sample, I don't scale the rest of the top velocity layer by RMS; I just multiply by the gain used to normalize the loudest strike, thus keeping the actual level differences. Then the RMS matching is used on the quieter velocity layers to match them to their respective ff samples. This method assumes that the top layer was recorded with uniform striking strength, and that the velocity-to-volume relationship is similar for all notes.
Last edited by 93143 on Thu Nov 16, 2006 11:24 pm, edited 1 time in total.
-
- KVRAF
- 1714 posts since 14 Mar, 2003 from Israel
Michiel Post (postpiano.com) mentions somewhere on his site using such a device to sample some of his piano libraries.deastman wrote:I wish there were an easy way of automating the sampling for multiple velocity layers. My parents have a beautiful 1928 Steinway B grand piano, and I'd love to make a free sample set of it. Thing is, I'm not going to bother doing a half-assed job of it, attempting to record a bunch of velocity layers by hand.
I remember when my parents were shopping for pianos, they mentioned that the serious stores have something called a "beater" which will sit on top of the keyboard and physically hammer on every key for a while to test for problems. If there was something like that which could be controlled from the computer and strike the keys with varying velocity values, I'd be set! Of course, I'm sure such a device, if it already exists, costs a lot of money...
CubaseStudio4 µTonic/Rapture Nitro/GS-201/Ohmicide/TBK 1&3
-
- KVRAF
- 7540 posts since 7 Aug, 2003 from San Francisco Bay Area
Cool! I'll check it out.spacefox wrote:Michiel Post (postpiano.com) mentions somewhere on his site using such a device to sample some of his piano libraries.deastman wrote:I wish there were an easy way of automating the sampling for multiple velocity layers. My parents have a beautiful 1928 Steinway B grand piano, and I'd love to make a free sample set of it. Thing is, I'm not going to bother doing a half-assed job of it, attempting to record a bunch of velocity layers by hand.
I remember when my parents were shopping for pianos, they mentioned that the serious stores have something called a "beater" which will sit on top of the keyboard and physically hammer on every key for a while to test for problems. If there was something like that which could be controlled from the computer and strike the keys with varying velocity values, I'd be set! Of course, I'm sure such a device, if it already exists, costs a lot of money...
Incomplete list of my gear: 1/8" audio input jack.
-
- KVRian
- 859 posts since 14 Sep, 2004
I've been working on a sampled studio piano myself. For what it's worth:
An alternative to either looping or having full length samples is to record about 20 seconds of each note. Almost no one needs more than that. (Adjust the sustain level and final decay with ADSR in the sampler.) This method lets you record all of the samples much faster. 60 keys,say, per layer X 20 seconds means that you could tentatively record a test layer in less than an hour.
I would avoid entirely a mechanical device: this would only give you the same amount of velocity for each strike level, which may sound desirable, but will not be to your advantage: as one plays, one is always adjusting the strength with which one hits each note to achieve a given sound. In an older piano, you will more than likely get very different volumes\amps from different keys struck with the same velocity, and will then have to go in and adjust all of the levels. I would create the velocity layers by ear and by touch if you're doing 4 or so velocities. (See below.) This concern with a mechanical device also lets you postpone the inevitable: setting up the mics, rolling up your sleeves, and sitting there while being entirely silent and pressing each key.
Other things to consider:
Consider sampling mainly the black keys: most people on older home pianos only play in a few keys (C and G, etc) so the felts may be in much better shape for the black keys. (Particularly around middle C.)
Listen to the individual piano notes carefully without trying to record. Determine what sound you want to reproduce. Make it
your goal to capture that exact sound, note for note, instead of trying to create a piano.
A good tuning is very important. One of the most common complaints about piano libraries is the tuning. A good tuning is important for the obvious reason, but will also give much better sounding notes if you stretch samples and will make sure that the fundamental sounds fully.
Record all of the samples on the same day or two, so the timbre doesn't change because of humidity, accidentally moving a mic a quarter of an inch, etc.
Take some time considering the scale of the timbre--something too often neglected: How bright and metallic will be your hardest strike, as comparared to your softest strike? Very hard to do 4 layers or so and get this right. Most sample sets assume a desire to get the hardest strike possible, and then end up with soft strikes that are too hard, so the transitions from the soft to the medium to the hardest strikes aren't abrupt. Going in the opposite direction can also be a problem: in trying to get the softest sounds right, you can end up neglecting the hardest strikes (which is better, to my ears, than neglecting the soft strikes, but it depends on what you want to play.)
Mic position is the hardest thing to determine, but you have to do it by means of experimentation. I think most of the time has to be spent here. Try everything that everyone suggests, but of course you finally have to go by your ears. A very big decision lies in where you put the mics in relation to the ends of the strings: pointing a mic near either end of a set of strings will catch more noise and rattle--a brighter sound. Putting the mics near the middle of the strings will give you more fundamental and string harmonics, but create the obvious problem of being so close to where the hammer strikes. In other words, pay close attention to where the hammer strikes the string--experiment with placing the mic halfway between the place at which the hammer strikes and the end of the string. Try minute adjustments from there. An inch can mean a huge difference in how much the hammer as opposed to the note is sounded. You may want to adjust the mics so you are always near this middle position between the point of hammer contact and where the string connects to the harp. If making this adjustment, be careful to move the entire mic and not just the head. The timbre will vary slightly if the mic faces the side of the string as opposed to the top or bottom: imagine the initial hammer strike as like, well, a hammer hitting the top of a pool of water: the energy spreads out differently from the sides as opposed to above or below where the hammer hits the top of the pool: much more force in the first huge ripple of water right around the strike and below it.
Another big decision in doing an upright is where in the room to put the piano: much of the sound of an upright comes from the wall it is placed against. It acts as a soundboard. If you move the piano to the middle of a room, you will only be recording the sound of the strings and hammers instead of the wall that both resonates and pushes the sound towards you as you sit and play. You can go back in and add convolution sounds, but the actual sound that may have made you want to record this piano won't be there.
To get the velocities by ear: record each velocity at a time, instead of all of the velocities of each note. This way, you can better determine that you are hitting the keys with about the same force, or at least getting a similar volume, for each layer. If your preamp has an amp meter, the combination of your ear and the visual feedback can let you get close to having the volume you want for each layer.
Pause every so often to listen in headphones and with monitors to the samples you've just recorded. Play the note on the piano, and then listen to the sound of the sample. Are you getting the sound you want? Better to check this way than to do 60 samples and then discover that the sound isn't what you want.
From what I gather, many people sampling a piano divide up their time strangely, in part because they are sampling in a hall, and then have to go home to map the sounds. As a result, they may record all of the samples and then go off to map them. Since you have the piano always available, I would try a different time line:
Once you've recorded an entire layer, map it out in fifteen minutes in Giga or Kontakt or whatever you're using. Start with the softest layer. Do a simple mapping that lets each note stand at all of the velocity layers. Play that softest layer on your midi keyboard and make whatever adjustments you can. Then return to the instrument and rerecord the samples that haven't worked. Then take them to the map. Do this until you have a layer right. Then back to the sampling, and then very quickly map the next layer, etc.
Be prepared to spend a lot of time on just a few samples. You may have to record three or four notes many, many times to get the sound you want. The area around middle C is often the hardest, partly because of mic placement and partly because the felts may be worn the most here. And because we listen so closely to this area.
The program Goldwave is great for working with piano samples: it has a section that lets you see the strengths\peaks of all of the harmonics in each sample, so you can double check yourself to be sure that the fundamental is coming through in each note. (Not hard to find a map on the internet listing all of the frequencies that are in each piano note, and thus determining if the fundamental is coming through enough.) I suggest using it, particularly if you are recording directly into the a computer: you can record into Goldwave and immediately see the peaks for each frequency. In any case, recording directly to the pc lets you keep Kontakt, etc, open and the recording software open at the same time, so you can listen to the sound you just recorded, and if you like it you can quickly map it by pressing alt-tab to go to Kontakt, etc.
Recording into a wave editor like Goldwave has another huge advantage: you can easily record many sounds as different "tracks" and then listen to each to compare the volume\timbre, etc. If you instead record in a stream, so each note comes after each other in a long recording, you must instead do a lot of cutting and pasting, and must fast forward and back to compare notes. You'll still have to trim note starts in Goldwave, but at least you won't have to cut up all of the big stream into parts for each note. Can save you a few hours.
Record at the highest sampling rate you can. The sound of each note may not seem so different between 44 and 96, but the result when you play chords is very noticable. (The Black Grand demo is helpful, here.)
All the best. Roll up those sleeves and get those mics out.
An alternative to either looping or having full length samples is to record about 20 seconds of each note. Almost no one needs more than that. (Adjust the sustain level and final decay with ADSR in the sampler.) This method lets you record all of the samples much faster. 60 keys,say, per layer X 20 seconds means that you could tentatively record a test layer in less than an hour.
I would avoid entirely a mechanical device: this would only give you the same amount of velocity for each strike level, which may sound desirable, but will not be to your advantage: as one plays, one is always adjusting the strength with which one hits each note to achieve a given sound. In an older piano, you will more than likely get very different volumes\amps from different keys struck with the same velocity, and will then have to go in and adjust all of the levels. I would create the velocity layers by ear and by touch if you're doing 4 or so velocities. (See below.) This concern with a mechanical device also lets you postpone the inevitable: setting up the mics, rolling up your sleeves, and sitting there while being entirely silent and pressing each key.
Other things to consider:
Consider sampling mainly the black keys: most people on older home pianos only play in a few keys (C and G, etc) so the felts may be in much better shape for the black keys. (Particularly around middle C.)
Listen to the individual piano notes carefully without trying to record. Determine what sound you want to reproduce. Make it
your goal to capture that exact sound, note for note, instead of trying to create a piano.
A good tuning is very important. One of the most common complaints about piano libraries is the tuning. A good tuning is important for the obvious reason, but will also give much better sounding notes if you stretch samples and will make sure that the fundamental sounds fully.
Record all of the samples on the same day or two, so the timbre doesn't change because of humidity, accidentally moving a mic a quarter of an inch, etc.
Take some time considering the scale of the timbre--something too often neglected: How bright and metallic will be your hardest strike, as comparared to your softest strike? Very hard to do 4 layers or so and get this right. Most sample sets assume a desire to get the hardest strike possible, and then end up with soft strikes that are too hard, so the transitions from the soft to the medium to the hardest strikes aren't abrupt. Going in the opposite direction can also be a problem: in trying to get the softest sounds right, you can end up neglecting the hardest strikes (which is better, to my ears, than neglecting the soft strikes, but it depends on what you want to play.)
Mic position is the hardest thing to determine, but you have to do it by means of experimentation. I think most of the time has to be spent here. Try everything that everyone suggests, but of course you finally have to go by your ears. A very big decision lies in where you put the mics in relation to the ends of the strings: pointing a mic near either end of a set of strings will catch more noise and rattle--a brighter sound. Putting the mics near the middle of the strings will give you more fundamental and string harmonics, but create the obvious problem of being so close to where the hammer strikes. In other words, pay close attention to where the hammer strikes the string--experiment with placing the mic halfway between the place at which the hammer strikes and the end of the string. Try minute adjustments from there. An inch can mean a huge difference in how much the hammer as opposed to the note is sounded. You may want to adjust the mics so you are always near this middle position between the point of hammer contact and where the string connects to the harp. If making this adjustment, be careful to move the entire mic and not just the head. The timbre will vary slightly if the mic faces the side of the string as opposed to the top or bottom: imagine the initial hammer strike as like, well, a hammer hitting the top of a pool of water: the energy spreads out differently from the sides as opposed to above or below where the hammer hits the top of the pool: much more force in the first huge ripple of water right around the strike and below it.
Another big decision in doing an upright is where in the room to put the piano: much of the sound of an upright comes from the wall it is placed against. It acts as a soundboard. If you move the piano to the middle of a room, you will only be recording the sound of the strings and hammers instead of the wall that both resonates and pushes the sound towards you as you sit and play. You can go back in and add convolution sounds, but the actual sound that may have made you want to record this piano won't be there.
To get the velocities by ear: record each velocity at a time, instead of all of the velocities of each note. This way, you can better determine that you are hitting the keys with about the same force, or at least getting a similar volume, for each layer. If your preamp has an amp meter, the combination of your ear and the visual feedback can let you get close to having the volume you want for each layer.
Pause every so often to listen in headphones and with monitors to the samples you've just recorded. Play the note on the piano, and then listen to the sound of the sample. Are you getting the sound you want? Better to check this way than to do 60 samples and then discover that the sound isn't what you want.
From what I gather, many people sampling a piano divide up their time strangely, in part because they are sampling in a hall, and then have to go home to map the sounds. As a result, they may record all of the samples and then go off to map them. Since you have the piano always available, I would try a different time line:
Once you've recorded an entire layer, map it out in fifteen minutes in Giga or Kontakt or whatever you're using. Start with the softest layer. Do a simple mapping that lets each note stand at all of the velocity layers. Play that softest layer on your midi keyboard and make whatever adjustments you can. Then return to the instrument and rerecord the samples that haven't worked. Then take them to the map. Do this until you have a layer right. Then back to the sampling, and then very quickly map the next layer, etc.
Be prepared to spend a lot of time on just a few samples. You may have to record three or four notes many, many times to get the sound you want. The area around middle C is often the hardest, partly because of mic placement and partly because the felts may be worn the most here. And because we listen so closely to this area.
The program Goldwave is great for working with piano samples: it has a section that lets you see the strengths\peaks of all of the harmonics in each sample, so you can double check yourself to be sure that the fundamental is coming through in each note. (Not hard to find a map on the internet listing all of the frequencies that are in each piano note, and thus determining if the fundamental is coming through enough.) I suggest using it, particularly if you are recording directly into the a computer: you can record into Goldwave and immediately see the peaks for each frequency. In any case, recording directly to the pc lets you keep Kontakt, etc, open and the recording software open at the same time, so you can listen to the sound you just recorded, and if you like it you can quickly map it by pressing alt-tab to go to Kontakt, etc.
Recording into a wave editor like Goldwave has another huge advantage: you can easily record many sounds as different "tracks" and then listen to each to compare the volume\timbre, etc. If you instead record in a stream, so each note comes after each other in a long recording, you must instead do a lot of cutting and pasting, and must fast forward and back to compare notes. You'll still have to trim note starts in Goldwave, but at least you won't have to cut up all of the big stream into parts for each note. Can save you a few hours.
Record at the highest sampling rate you can. The sound of each note may not seem so different between 44 and 96, but the result when you play chords is very noticable. (The Black Grand demo is helpful, here.)
All the best. Roll up those sleeves and get those mics out.
-
- KVRAF
- 7540 posts since 7 Aug, 2003 from San Francisco Bay Area
Jake,Jake Jackson wrote: All the best. Roll up those sleeves and get those mics out.
Thanks for all the good advice. I'm feeling inspired to get started on a sample set for my parents Steinway B, as well as my own crappy Baldwin Spinet... just as soon as I have a few other things out of the way, such as having a new baby in the next week or two, and then finding a few hours of free, quiet time to record. Yeah, right...
I had one other thought... what do you think about recording a whole bunch of strikes for one note at different velocity levels, say 16 strikes per note, and then moving on to the next note? It might take a while to match up a good set of velocity layers across the keyboard, but at least I could get a lot of sampling done quickly, and then worry about the mapping later. Just a thought...
As for microphones, my collection is very limited. The best thing I've got is a Neumann TLM-103, although I could borrow a couple of U87's from a friend. Would these be appropriate for piano sampling, or are they too bright?
Also, my Baldwin is in a large, open, multi-level townhouse. I thought it might be fun to record a set of samples from way at the other end of the open space, to get a large ambient sound, and perhaps blend that with a close-mic'd version. I suppose a PZM would be ideal for that.
Incomplete list of my gear: 1/8" audio input jack.
-
- KVRAF
- 3645 posts since 27 Nov, 2003 from beach side australia
whoah i was going to sample my grans pianola but its just too much work .. i went in the vitaudio group buy instead
-
- KVRAF
- 16154 posts since 2 Dec, 2003 from Nashville, TN
I guess I could chime in here!
Most of what is going on right now is Watto's part. I'm mostly just mapping and editing and such. Personally, as a long time piano player, I believe that recording as many layers as possible is preferred. You don't have to use them all, but if you find you want more, it's harder to go back and get the same sound after you've torn everything down. Plus, it allows me to make different versions, and get as detailed as needed.
I think that only white keys would be fine, but it depends on time. Of course, if you have the time, record everything. The advantage to having that is because if I get to editing and find that C2 has some wierd sound in it on one of the samples, I can just use C# or whatever instead. It will allow me to have options when mapping, to avoid having the above happen, and also to avoid the sampler having to do any more pitch changing than it has to.
Ok, gotta go. I've been so busy the last few days, but I'll monitor this thread. I'm also open to suggestions on mapping. Cheers!
Brent
Most of what is going on right now is Watto's part. I'm mostly just mapping and editing and such. Personally, as a long time piano player, I believe that recording as many layers as possible is preferred. You don't have to use them all, but if you find you want more, it's harder to go back and get the same sound after you've torn everything down. Plus, it allows me to make different versions, and get as detailed as needed.
I think that only white keys would be fine, but it depends on time. Of course, if you have the time, record everything. The advantage to having that is because if I get to editing and find that C2 has some wierd sound in it on one of the samples, I can just use C# or whatever instead. It will allow me to have options when mapping, to avoid having the above happen, and also to avoid the sampler having to do any more pitch changing than it has to.
Ok, gotta go. I've been so busy the last few days, but I'll monitor this thread. I'm also open to suggestions on mapping. Cheers!
Brent
My host is better than your host
-
- KVRian
- 859 posts since 14 Sep, 2004
deastman:
About recording all of the velocities of one key: I wouldn't, based on my own experience: you would then have to match up all of the other velocities of all the other keys to that one key, which could be a major headache. You would actually get a playable, but limited, sound faster by doing one layer at a time. Once the mics are to your liking, you could do a layer in an hour if you record directly into the computer, even if you stop to compare the sample to playing the note live.
I agree that in general the more keys and the more velocities the better. Maybe I misunderstood the original post--I thought the goal was a first try at some home sampling that wouldn't take too long.
In any case, I would spend most of the time on:
1. Playing with the mics to get the sound you want. (Which may only take a few minutes--we've all had the experience of getting them set up fast.)
2. Listening to the sound of a note before recording it (and then comparing the recording to the note played live). Much better to slowly record 4 layers that sound excellent than quickly do 10 that are decent. Best to do 10 excellent layers, of coure, but then you get into months of work, losing the day job, etc.
About the velocity layering: One problem. to my ears, with many of the sample libraries I've heard is that the trebles are just much, much too low in volume. On a real piano, the treble notes require a harder strike to sound at the same volume as a bass note. Shouldn't be a problem--for the sake of realistic playing, one should be able to keep the real velocity layers and just let the player play harder on the treble notes. But somehow, in the sample libraries, the timbre gets thinner faster than it should with harder strikes. The sound does get more metallic on a real piano (particularly a Yamaha), but the fundamental doesn't get lost as fast. More generally, I find this to be a problem with a note or two on many sampled pianos--the sound that is recorded leaves out the fundamental.
I hope things go well for you. (And that you'll post a few mp3's of a layer or two here.)
About recording all of the velocities of one key: I wouldn't, based on my own experience: you would then have to match up all of the other velocities of all the other keys to that one key, which could be a major headache. You would actually get a playable, but limited, sound faster by doing one layer at a time. Once the mics are to your liking, you could do a layer in an hour if you record directly into the computer, even if you stop to compare the sample to playing the note live.
I agree that in general the more keys and the more velocities the better. Maybe I misunderstood the original post--I thought the goal was a first try at some home sampling that wouldn't take too long.
In any case, I would spend most of the time on:
1. Playing with the mics to get the sound you want. (Which may only take a few minutes--we've all had the experience of getting them set up fast.)
2. Listening to the sound of a note before recording it (and then comparing the recording to the note played live). Much better to slowly record 4 layers that sound excellent than quickly do 10 that are decent. Best to do 10 excellent layers, of coure, but then you get into months of work, losing the day job, etc.
About the velocity layering: One problem. to my ears, with many of the sample libraries I've heard is that the trebles are just much, much too low in volume. On a real piano, the treble notes require a harder strike to sound at the same volume as a bass note. Shouldn't be a problem--for the sake of realistic playing, one should be able to keep the real velocity layers and just let the player play harder on the treble notes. But somehow, in the sample libraries, the timbre gets thinner faster than it should with harder strikes. The sound does get more metallic on a real piano (particularly a Yamaha), but the fundamental doesn't get lost as fast. More generally, I find this to be a problem with a note or two on many sampled pianos--the sound that is recorded leaves out the fundamental.
I hope things go well for you. (And that you'll post a few mp3's of a layer or two here.)
-
- KVRian
- 500 posts since 13 Oct, 2004 from Durham, NC USA
I agree with everything Jake says except I would encourage pulling the piano and miking it from behind, since the piano's soundboard is exposed that way.
My first addition is to forget about looping when sampling: that's done later. As Jake says, record long samples, but feel free to cut off at 20 or 30 seconds for the low notes that last that long. (Note that only low notes last that long!)
IMHO, 3 or 4 layers is a minimum for a good result, and more is better -- but each layer adds considerable work. However, note that timbre varies a lot more at lower levels than higher ones, so all layers needn't go the full length.
I recommend you start by doing a partial sample, using only 2 or 3 layers and only 2 notes per octave, and plan to do a second attempt later. You'll learn from the first attempt and the results. One very important thing you'll learn is whether your mike position is good for the range of the piano.
The job will be made incredibly easier using my soundfont tools. I'll be happy to help, btw. To use these tools, record each velocity layer in a separate wave file.
To record a velocity layer, a "thumper" would be best, but making a good multi-velocity thumper is a real engineering exercise. Here's what I do. I assume you have a 24-bit soundcard. Otherwise, don't bother.
1) Find the loudest note, and set record levels so it doesn't clip when played at the highest velocity you plan to record.
2) At regular intervals along the keyboard (say, 4 places), play the quietest and loudest velocities you plan to record, and make a note of the peak level. Do this on neighboring keys too, and pray that the levels are consistent. If they're not, you'll have trouble. Of course, this is difficult, since it's hard to play a single "FF" note on demand. Just do your best, but this is a critical part so really do your best.
3) You now have a set of peak levels for 4 or so points on the keyboard. Interpolating between them, you should be able to estimate the peak velocities you expect to get for the notes in between, for the highest and lowest layers. Set yourself up with a set of target velocities for each octave or so on the keyboard. BTW, when recording jRhodes3, I did this more or less in my head. "Close, but steady" is the goal, not mathmatical perfection.
4) Start the recorder and record a layer, watching the peak meters on your DAW. Each time you strike a note, if you didn't hit the desired peak level, drop it right away (you have at least a second or so, which is plenty of time. Keep trying on the same note until you get it right, and then hold it until it fades, or at most 20-30 sec. Then move on to the next note.
BTW, the above process would be better using RMS meters rather than peak meters. If only more DAWs had 'em! If you know of a VST plugin that offers easy to read RMS meters with a "peak RMS hold" function, go for it!
Don't mike too close (with one-point miking, as you plan to do) or the octaves closest to the mikes will be too loud. I'd guess that 4 feet away should be ideal.
PREPARE THE ROOM! Probably best to deaden it quite a bit, to avoid recording too much room along with the samples. It depends a lot on the quality and nature of the room: if it's in a nice big lodge with wood floors and high ceilings ... well, that's a great sound, especially for upright. But a reverby piano doesn't work so well live in combo situations, where venues already have too much natural reverb. Remember that the more room you get in the sound, the less versatile the result will be. But the more natural it might be, so it's a tradeoff, if the room sounds great.
For recording in stereo, I highly recommend trying mid-side technique, where one mike (mid) points directly at the piano, and the other (side) points roughly 90 degrees to one side (say, the left). Mike capsules should be as close as possible without touching. The two mikes don't need to be matched, or even the same type of mike. The side mike is only for stereo imagery. More on this in another post, if you're interested.
I need to collect this stuff into a web page. Might be a good one for a Wiki!
My first addition is to forget about looping when sampling: that's done later. As Jake says, record long samples, but feel free to cut off at 20 or 30 seconds for the low notes that last that long. (Note that only low notes last that long!)
IMHO, 3 or 4 layers is a minimum for a good result, and more is better -- but each layer adds considerable work. However, note that timbre varies a lot more at lower levels than higher ones, so all layers needn't go the full length.
I recommend you start by doing a partial sample, using only 2 or 3 layers and only 2 notes per octave, and plan to do a second attempt later. You'll learn from the first attempt and the results. One very important thing you'll learn is whether your mike position is good for the range of the piano.
The job will be made incredibly easier using my soundfont tools. I'll be happy to help, btw. To use these tools, record each velocity layer in a separate wave file.
To record a velocity layer, a "thumper" would be best, but making a good multi-velocity thumper is a real engineering exercise. Here's what I do. I assume you have a 24-bit soundcard. Otherwise, don't bother.
1) Find the loudest note, and set record levels so it doesn't clip when played at the highest velocity you plan to record.
2) At regular intervals along the keyboard (say, 4 places), play the quietest and loudest velocities you plan to record, and make a note of the peak level. Do this on neighboring keys too, and pray that the levels are consistent. If they're not, you'll have trouble. Of course, this is difficult, since it's hard to play a single "FF" note on demand. Just do your best, but this is a critical part so really do your best.
3) You now have a set of peak levels for 4 or so points on the keyboard. Interpolating between them, you should be able to estimate the peak velocities you expect to get for the notes in between, for the highest and lowest layers. Set yourself up with a set of target velocities for each octave or so on the keyboard. BTW, when recording jRhodes3, I did this more or less in my head. "Close, but steady" is the goal, not mathmatical perfection.
4) Start the recorder and record a layer, watching the peak meters on your DAW. Each time you strike a note, if you didn't hit the desired peak level, drop it right away (you have at least a second or so, which is plenty of time. Keep trying on the same note until you get it right, and then hold it until it fades, or at most 20-30 sec. Then move on to the next note.
BTW, the above process would be better using RMS meters rather than peak meters. If only more DAWs had 'em! If you know of a VST plugin that offers easy to read RMS meters with a "peak RMS hold" function, go for it!
Don't mike too close (with one-point miking, as you plan to do) or the octaves closest to the mikes will be too loud. I'd guess that 4 feet away should be ideal.
PREPARE THE ROOM! Probably best to deaden it quite a bit, to avoid recording too much room along with the samples. It depends a lot on the quality and nature of the room: if it's in a nice big lodge with wood floors and high ceilings ... well, that's a great sound, especially for upright. But a reverby piano doesn't work so well live in combo situations, where venues already have too much natural reverb. Remember that the more room you get in the sound, the less versatile the result will be. But the more natural it might be, so it's a tradeoff, if the room sounds great.
For recording in stereo, I highly recommend trying mid-side technique, where one mike (mid) points directly at the piano, and the other (side) points roughly 90 degrees to one side (say, the left). Mike capsules should be as close as possible without touching. The two mikes don't need to be matched, or even the same type of mike. The side mike is only for stereo imagery. More on this in another post, if you're interested.
I need to collect this stuff into a web page. Might be a good one for a Wiki!
-
- KVRian
- 500 posts since 13 Oct, 2004 from Durham, NC USA
And do not do any gain adjustments! Also, no normalization. Any normalization to be done should only be done to build the soundfont, not to the original sample sets. While I prefer normalized layers in compiled soundfonts, this is done as a subsequent processing step. Normalizing before giving the files to the person doing the tech work is thowing away important information.
An alternative, which can be used if the noise is too high in the quietest layers, is to adjust the gain before recording those layers, but record the exact gain change!
An alternative, which can be used if the noise is too high in the quietest layers, is to adjust the gain before recording those layers, but record the exact gain change!
-
- KVRian
- 859 posts since 14 Sep, 2004
I'm worried that we'll have him so anxious, with so many things to consider, that he'll never get this started. Go for it. Get two layers done one afternoon and you'll have the motivation to record the others.
But I do want to add that I agree about taking ten minutes to damp the room with blankets, etc. A dryer sound is better. Let in the reverberations of the sounds inside the piano, but not the room reflections. Or if you like the sound of the room for everyday playing, make two different sample sets. (Yikes?)
But I do want to add that I agree about taking ten minutes to damp the room with blankets, etc. A dryer sound is better. Let in the reverberations of the sounds inside the piano, but not the room reflections. Or if you like the sound of the room for everyday playing, make two different sample sets. (Yikes?)