Can Loudspeakers Accurately Reproduce The Sound Of Real Instruments...and Do You Care?

[Theo]

What's interesting about the FT approach is that it's not fully invertible.

You're right. The Fourier transform is actually a complex function. This means that inversion must use phase as well as frequency in order to recover the exact input function. Otherwise, the input signal is supposed to be minimum phase to be reconstructed accurately (if I remember correctly...).

 

[bennetng]

The talk about timing of real instruments playing same or different notes...

20-30 years ago there are something like this:
https://en.wikipedia.org/wiki/Roland_SC-55

In fact, if you play a MIDI file in Windows, the sound set is from SC-55 or its variant:
https://sound.stackexchange.com/que...-the-default-windows-8-midi-soundfont-located

It is a ~4MB file and it contains hundreds of sampled instruments.

But nowadays it is something like this:
http://www.soundsonline.com/pianos
It is 57GB (gold version) and 282GB (platinum version), with only piano sounds.

Now you may think something so big must be 24/192 and so on... No. I have the gold version and it only uses 16/44 sample, but still 57GB!

I made a transcription of this music:

Here is my version, with that piano library. Yes I know some notes, chords etc are different but they are in the musical aspect rather than audio.
https://hydrogenaud.io/index.php?action=dlattach;topic=114240.0;attach=11266

Why these libraries are so huge? Among other reasons, it captures many different single-note (not loop, chord or phrase) playback technique, even same techniques are captured multiple times. One reason that people can easily discern real vs synthesized instrument is because real instruments always produce different sounds even when using the same technique. If a virtual instrument always use the same sample all the time it is very easy to notice the repetition and it is irritating.

While it may be irrelevant to this thread, timing precision of MIDI is expressed in ticks per quarter note (TPQN), the highest limit is 32767, which means it can divide the timing of a quarter note into 32767 parts. However in general usage, 96-960 TPQN are common. Also, the absolute length of a quarter note depends on tempo, therefore the timing is also relative to tempo.

 

[andreasmaaan]

The talk about timing of real instruments playing same or different notes...

20-30 years ago there are something like this:
https://en.wikipedia.org/wiki/Roland_SC-55

In fact, if you play a MIDI file in Windows, the sound set is from SC-55 or its variant:
https://sound.stackexchange.com/que...-the-default-windows-8-midi-soundfont-located

It is a ~4MB file and it contains hundreds of sampled instruments.

But nowadays it is something like this:
http://www.soundsonline.com/pianos
It is 57GB (gold version) and 282GB (platinum version), with only piano sounds.

Now you may think something so big must be 24/192 and so on... No. I have the gold version and it only uses 16/44 sample, but still 57GB!

I made a transcription of this music:

Here is my version, with that piano library. Yes I know some notes, chords etc are different but they are in the musical aspect rather than audio.
https://hydrogenaud.io/index.php?action=dlattach;topic=114240.0;attach=11266

Why these libraries are so huge? Among other reasons, it captures many different single-note (not loop, chord or phrase) playback technique, even same techniques are captured multiple times. One reason that people can easily discern real vs synthesized instrument is because real instruments always produce different sounds even when using the same technique. If a virtual instrument always use the same sample all the time it is very easy to notice the repetition and it is irritating.

While it may be irrelevant to this thread, timing precision of MIDI is expressed in ticks per quarter note (TPQN), the highest limit is 32767, which means it can divide the timing of a quarter note into 32767 parts. However in general usage, 96-960 TPQN are common. Also, the absolute length of a quarter note depends on tempo, therefore the timing is also relative to tempo.

That's really interesting, thanks.

Did you sit down and transcribe the entire work by ear?!

I would add that another reason that midi-based reproductions tend to be discernible from the original is to do with reverberation.

Re: a virtual instrument always using the same sample, there are midi effects available in some DAWs (Ableton Live for example) that randomise things like velocity or timing within thresholds set by the user. I've found subtle application of these to be quite effective at overcoming this particular shortcoming, although I use them in the context of electronic music only.

 

[RayDunzl]

When I first got a windows PC, I couldn't figure out how to write code for it, Never did, gave up.

So I dropped back to command line and wrote a little C to create rule based tunes one sample at a time in 16/44.1.

Multi-voice, stereo, envelopes, modulation, reverb, microtonal scales, everything I could think of in the Hotel room...

I learned a lot about digital sound from that now ancient adventure.

Haven't fooled with it since.

 

[bennetng]

Did you sit down and transcribe the entire work by ear?!

Yes.

I would add that another reason that midi-based reproductions tend to be discernible from the original is to do with reverberation.

It depends on libraries. Some libraries are recorded with different mic positions so reverb, ambience and such are baked in. You may ask "what if I use other external libraries in the same mix, it sounds like a strange approach to me".
Well, in this case, either use another library from the same series, for example I used this orchestra in the previous post:
http://www.soundsonline.com/symphonic-orchestra

Or use dry instrument and fiddle with reverb plugins so the mix sounds more natural.

randomise things like velocity or timing within thresholds set by the user.

Sure, for example chorus plugin. But it sounds fake, for example, try to record a solo violin and randomize things to mimic a violin section with for example 10-15 violins, it will sound horrible for sure. In virtual instruments, they are recorded with a real violin section.

 

[andreasmaaan]

Yes.

Very impressive!

It depends on libraries. Some libraries are recorded with different mic positions so reverb, ambience and such are baked in. You may ask "what if I use other external libraries in the same mix, it sounds like a strange approach to me".
Well, in this case, either use another library from the same series, for example I used this orchestra in the previous post:
http://www.soundsonline.com/symphonic-orchestra

Or use dry instrument and fiddle with reverb plugins so the mix sounds more natural.

Totally agree.

Sure, for example chorus plugin. But it sounds fake, for example, try to record a solo violin and randomize things to mimic a violin section with for example 10-15 violins, it will sound horrible for sure. In virtual instruments, they are recorded with a real violin section.

Wouldn't a chorus effect be an audio effect rather than a midi effect?

I'm thinking about midi effects, which can randomise e.g. the midi attack velocities of a sequence.

Here's an example from Ableton Live:

 

[bennetng]

Wouldn't a chorus effect be an audio effect rather than a midi effect?

There are many different approaches to apply effects, it can be applied in a non-DSP way like randomizing the MIDI commands themselves rather than the rendered audio. The basic General MIDI standard also defines audio DSP effects (reverb/chorus) after the MIDI commands are converted to audio in each MIDI channel.
http://www.nortonmusic.com/midi_cc.html

The above is just an example, different software/hardware can change the implementation.

In case of modern DAWs, you can also route each rendered MIDI channel to the effect bus for additional processing.

In fact, unless you are just double clicking a MIDI file and listen to it, modern MIDI+DAW workflow rarely follows the original General MIDI standard.

A real SC-55, despite using a similar (tiny) sound set as the Windows synth, sounds lightyears better. It has a built-in audio DSP as well.

 

[andreasmaaan]

There are many different approaches to apply effects, it can be applied in a non-DSP way like randomizing the MIDI commands themselves rather than the rendered audio. The basic General MIDI standard also defines audio DSP effects (reverb/chorus) after the MIDI commands are converted to audio in each MIDI channel.
http://www.nortonmusic.com/midi_cc.html

The above is just an example, different software/hardware can change the implementation.

In case of modern DAWs, you can also route each rendered MIDI channel to the effect bus for additional processing.

In fact, unless you are just double clicking a MIDI file and listen to it, modern MIDI+DAW workflow rarely follows the original General MIDI standard.

A real SC-55, despite using a similar (tiny) sound set as the Windows synth, sounds lightyears better. It has a built-in audio DSP as well.

Yes, totally aware of all that! Was just making sure I'd made clear what I meant when I referred to "midi effects" (as distinct from audio effects) in my previous post, as the distinction seemed to have been lost

 

[Cosmik]

The talk about timing of real instruments playing same or different notes...

Why these libraries are so huge? Among other reasons, it captures many different single-note (not loop, chord or phrase) playback technique, even same techniques are captured multiple times.

Well with large samples (an early example being the famous Gigasampler) we're effectively playing a CD-quality recording of a piano. But it would be interesting to know if any piano music fans or piano players here feel that digital pianos are ever a convincing replacement for an acoustic one.

I am in the position of having recently got rid of our acoustic piano and replaced it with a modest digital one - a Kawai stage piano with fairly small speakers. And yet... as a player, I am not missing the acoustic one at all. But I think actively playing an instrument is probably quite a bit different from passively listening to one - your attention is not just on the sound.

If I hear the piano being played from the next room it certainly sounds like a piano. In the same room, I think it is lacking something. I have tried external speakers and a subwoofer. It's OK, but in that case, I noticed that the lack of mechanical coupling between the piano and speakers made it less satisfying to play. With the inbuilt speakers the piano vibrates in sympathy with the sound and you feel it with your fingers.

 

[andreasmaaan]

With the inbuilt speakers the piano vibrates in sympathy with the sound and you feel it with your fingers.

Time to build a new speaker into the piano?

 

[bennetng]

Well with large samples (an early example being the famous Gigasampler) we're effectively playing a CD-quality recording of a piano. But it would be interesting to know if any piano music fans or piano players here feel that digital pianos are ever a convincing replacement for an acoustic one.

I am in the position of having recently got rid of our acoustic piano and replaced it with a modest digital one - a Kawai stage piano with fairly small speakers. And yet... as a player, I am not missing the acoustic one at all. But I think actively playing an instrument is probably quite a bit different from passively listening to one - your attention is not just on the sound.

If I hear the piano being played from the next room it certainly sounds like a piano. In the same room, I think it is lacking something. I have tried external speakers and a subwoofer. It's OK, but in that case, I noticed that the lack of mechanical coupling between the piano and speakers made it less satisfying to play. With the inbuilt speakers the piano vibrates in sympathy with the sound and you feel it with your fingers.

Ah yes. Gigasampler. Nowadays people are all talking about ASIO and such, no one remembered GSIF.

I have a MIDI keyboard but I am not a pianist. I can only play some simple tunes in real time. At most time my MIDI works are done in mouse. I also learned Erhu in the past but the more I worked with MIDI the more I think mouse is my instrument.

 

[Cosmik]

Time to build a new speaker into the piano?

That's my plan

 

[MRC01]

...What instrument would you say is the most difficult for your speakers to reproduce with realism? ...

Another vote for piano here. Also:
harp: incredibly deep yet soft bass with complex timbre, with super HF transient plucks.
drums (bass, tympani): to capture not only the tone, but also the dynamic attack when struck
bagpipes: of various sizes, timbres with lots of HF energy, usually sound artificial with presence emphasis and rolled off HF
voices: rare to capture it well, without artificial presence and HF emphasis
Problem is, these are often poorly recorded so during playback it can be hard to differentiate the root case of problems: recording, or playback.

What is your opinion of the digital pianos that proliferate? ...

They don't sound like pianos. The best ones I've heard sound like a half-decent recording of a piano.

...
You need good power for a believable piano.
.. [snipped audiophile porn photos] ...
About 275W idle power draw, so, it ain't green...

The Adcom 5800 I use is similar. Draws 250W idle, 1800VA at full tilt, and heats the room. But you need that kind of power to control the speakers and get clean effortless sounding full dynamics.

 

[Cosmik]

They don't sound like pianos. The best ones I've heard sound like a half-decent recording of a piano.

One thing I've noticed, though, is that some new acoustic upright pianos sound very 'loud' and, to me, artificial and characterless. I wonder if they'll start making acoustic pianos sound more like digital.

 

[andreasmaaan]

Piano tends to be most revealing of nonlinear distortion in audibility studies I’ve read.

Castanets are often used in phase distortion audibility studies on the basis that impulses best reveal phase distortion and castanets best resemble impulses. Not sure if there’s actually an evidential basis for this however.

 

[MRC01]

...
Castanets are often used in phase distortion audibility studies on the basis that impulses best reveal phase distortion and castanets best resemble impulses. Not sure if there’s actually an evidential basis for this however.

There is something special about castanets. When testing low pass filters at high threshold frequencies, castanets make the highest frequency thresholds easier to hear. An FFT of a good recording of castanets has energy to and beyond 20 kHz, similar to jangling keys in front of the mic.

 

[andreasmaaan]

There is something special about castanets. When testing low pass filters at high threshold frequencies, castanets make the highest frequency thresholds easier to hear. An FFT of a good recording of castanets has energy to and beyond 20 kHz, similar to jangling keys in front of the mic.

Yeh, also the most popular instrument for testing things like sample rates and anti-imaging filters etc for this reason.

 

[MRC01]

One thing I've noticed, though, is that some new acoustic upright pianos sound very 'loud' and, to me, artificial and characterless. I wonder if they'll start making acoustic pianos sound more like digital.

Ask any musician, but especially piano players, which acoustic pianos they like and dislike. Oh boy that's a touchy subject! I've seen less contentious discussions about religion and politics.

 

[oivavoi]

One thing I've noticed, though, is that some new acoustic upright pianos sound very 'loud' and, to me, artificial and characterless. I wonder if they'll start making acoustic pianos sound more like digital.

I think this observation is correct. Modern pianos - much like speakers by the way - are often made to sound good at show rooms. In order to stand out of the pack they are made to be very loud and be overly bright. This may be impressive at first, but it's not ideal for the home environment and long term playing. Furthermore, modern pianos are often tuned according to a "linear" philosophy. This is a good thing in sound reproduction, but not necessarily in an acoustic instrument. What I mean by that is that the different registries - bass, mid, treble - are tuned to have a similar character - to be very clear and distinct. By contrast, the older philosophy was to tune the bass to be "fat"; and the mid and treble to be clean and clear. When using the piano as a solo instrument, this is arguably better. This way you get a warm feel from the fat bass, but you also get a sense of clarity and singing from the higher registries. This mimics an actual orchestra, where the different instruments have tonal characters which complement each other.

My favorite pianos are often older pianos which have been completely rebuilt and restored. I like the older tonal philosophy, and I must admit that I also really like playing on ivory (my apologies, all ye endangered ivory-carrying species of the world).

Btw, I agree with @MRC01 that digital pianos don't sound or feel like acoustic pianos to me. I've played the piano daily since I was six. Some years ago I really wanted to think that the digitals were there, because it would make life so much easier. But after having lived with a digital piano for some years now, and tried out all of the high-end digitals, I have come to the sad conclusion that it's just not there yet. There are several reasons for this. First and foremost it's about the built-in speakers in the DPs themselves, which usually suck, and don't disperse sound the same way an acoustic piano does. Secondly it's about how an acoustic instrument responds to different combinations of speed, touch, chord progressions, input, etc. It's so complex that it's almost impossible to get right with sampling and digital input.

The closest I get to a real piano experience from playing digital pianos is actually not through sampling, but through a virtual modelled piano called Pianoteq. https://www.pianoteq.com
I find that this modelled piano gets closer to how a real piano behaves than the sampled pianos do. I also prefer playing with headphones, because a) the internal speakers of all DPs are bad, b) using external monitors becomes artificial, somehow.

That said, there are people who don't feel this way, and think that DP's currently can rival acoustics. It may partly be about what one is used to, I think.

Ok, piano rant over!

 

[svart-hvitt]

I think this observation is correct. Modern pianos - much like speakers by the way - are often made to sound good at show rooms. In order to stand out of the pack they are made to be very loud and be overly bright. This may be impressive at first, but it's not ideal for the home environment and long term playing. Furthermore, modern pianos are often tuned according to a "linear" philosophy. This is a good thing in sound reproduction, but not necessarily in an acoustic instrument. What I mean by that is that the different registries - bass, mid, treble - are tuned to have a similar character - to be very clear and distinct. By contrast, the older philosophy was to tune the bass to be "fat"; and the mid and treble to be clean and clear. When using the piano as a solo instrument, this is arguably better. This way you get a warm feel from the fat bass, but you also get a sense of clarity and singing from the higher registries. This mimics an actual orchestra, where the different instruments have tonal characters which complement each other.

My favorite pianos are often older pianos which have been completely rebuilt and restored. I like the older tonal philosophy, and I must admit that I also really like playing on ivory (my apologies, all ye endangered ivory-carrying species of the world).

Btw, I agree with @MRC01 that digital pianos don't sound or feel like acoustic pianos to me. I've played the piano daily since I was six. Some years ago I really wanted to think that the digitals were there, because it would make life so much easier. But after having lived with a digital piano for some years now, and tried out all of the high-end digitals, I have come to the sad conclusion that it's just not there yet. There are several reasons for this. First and foremost it's about the built-in speakers in the DPs themselves, which usually suck, and don't disperse sound the same way an acoustic piano does. Secondly it's about how an acoustic instrument responds to different combinations of speed, touch, chord progressions, input, etc. It's so complex that it's almost impossible to get right with sampling and digital input.

The closest I get to a real piano experience from playing digital pianos is actually not through sampling, but through a virtual modelled piano called Pianoteq. https://www.pianoteq.com
I find that this modelled piano gets closer to how a real piano behaves than the sampled pianos do. I also prefer playing with headphones, because a) the internal speakers of all DPs are bad, b) using external monitors becomes artificial, somehow.

That said, there are people who don't feel this way, and think that DP's currently can rival acoustics. It may partly be about what one is used to, I think.

Ok, piano rant over!

Another perspective on new vs old

https://www.sciencemag.org/news/2017/05/million-dollar-strads-fall-modern-violins-blind-sound-check