Toward High-Quality Live Sound for Digital Piano

Keyboardists often complain that their digital/electronic piano sounds harsh, honky, thinmuffled, resonant, tinny, strident, boxy, like a toy/budget instrument, and other kinds of terrible and awful, such as the “ice-pick-in-the-ear effect”. I’ve been frustrated by these issues myself; no one feels inspired playing an instrument that sounds bad!

After a few years of working on understanding the causes of these problems, I’m convinced they can be solved by careful loudspeaker design, especially with attention to directivity control. It is possible to achieve a much more inspiring sound with existing keyboards. However, most off-the-shelf sound systems (e.g. “keyboard” amps, monitors and compact PAs) aren’t up to the task, for reasons described below.

Update: based on my investigations here and elsewhere, I’ve designed a new loudspeaker system to deal with the challenges of turning piano samples into naturalistic live piano sound. Visit taylorsoundlabs.com to purchase or get more info.

Headphones Sound Good

Importantly, many players are happy with the sound of their piano in headphones — only their live sound causes grief. (Here “live” can mean anything from a public performance to playing by yourself at home.) The typical experience is described in comments like:

“Why do these digital pianos sound so good through headphones but kinda suck when listening through speakers?” [Busch on forums.musicplayer.com]

“I have a pair of fairly high end speakers … and the sound plus the overall playing experience is nowhere near the same compared to my $99 set of phones.” [Dave on forums.musicplayer.com]

“I [tried out] the supposedly best options in the way of keyboard amps … and was shocked and confused to find that the sound of this equipment (in the thousands of dollars) for naturalistic piano was completely unusable … Seems like people trying for naturalistic sound from dig pianos in smaller clubs sound tinny. Why should it be so hard to get acceptable sounds from a dig piano anyway, why such a discrepancy between the headphones and what’s coming through the amp??” [ziggy on gearslutz.com]

This means that sound problems don’t originate in the piano samples themselves, but are instead an artefact of the live sound system. The quality of piano samples does vary — some are clearly better than others — but the fact that most players are happy with their sound in headphones shows that piano samples themselves are not the root cause of players’ frustration.

Causes of Bad Live Piano Sound

Instead, the evidence I’ve gathered (by spectral analysis of the signals at my digital piano’s outputs and acoustic measurements of several amps and PAs) points to the following causes of bad live sound:

  • Phase cancellation. Phase differences between left and right channels cause phase cancellation (comb filtering) when the channels are summed, either electronically at a summed-to-mono output (as is typical of the L-MONO output on most keyboards) or acoustically at the listener’s ears when playing in “stereo”. This results in a characteristic thin, tinny, boxy, resonant, toy-instrument sound, often similar to a phasing or flanging effect. I’ve posted some measurements demonstrating this effect here and here. This post also has some good discussion. (Incidentally, this also explains why these pianos sound good in headphones: there is no left-right summing in ‘phones, hence no chance of phase cancellation.)
  • Polyphonic interference. When playing a chord, different notes will typically generate harmonics that share a common frequency. Phase differences between these harmonics can lead to phase cancellation, depending on small differences in timing (on the order of milliseconds) between the struck notes. This causes random (sometimes unpleasant) changes in timbre from one sounding of a chord to the next. Measurements demonstrating this effect are shown in this post.
  • Loudspeaker resonances. Unless great care is taken in loudspeaker design, there will be audible resonances due to e.g. standing waves in the cabinet or in the throat of a horn, so some pitches (or their harmonics) will sound much louder than others. Live piano seems to be especially revealing of these resonances, which are the cause of the harsh, strident, “ice-pick-in-the-ear” effects that many players complain of. Most common seems to be a resonance around 1kHz that gives a piercing quality to a few notes around high C.

Problems with Keyboard Amps / PAs

With the exception of polyphonic interference, these causes of bad piano sound are actually made worse by the typical approach to design of sound systems for live performance. Floyd Toole summarizes the prevailing design goals (my emphasis added):

Directional control is critical in designing sound reinforcement systems, the purpose of which is to deliver sound to the audience without exciting excessive reflections and reverberation within the room itself. The challenge is to put more of the audience in a predominantly direct sound field, precisely the opposite of a live concert hall experience. [1, p.47]

These criteria are great for speech intelligibility and for performance in large venues, but they contribute to bad piano sound in at least three ways.

First, and most obviously, a piano invariably sounds too “dry” when played through an amp / PA without reverb, because the sound system provides very little in the way of room reflections and natural reverberation — quite unlike an acoustic piano. With acoustic piano we actually do want to approximate a live experience. This changes the design goal entirely, favoring relatively high levels of reverberant sound and early lateral reflections [1] — precisely the opposite of what typical keyboard amps and PAs aim to do. Adding synthetic reverb helps, but it often sounds unnatural: the human ear is pretty good at detecting the mismatch between synthetic reverb and the natural reverberation of a room.

More importantly, and unfortunately for stereo piano, a high level of direct sound presents a worst-case acoustical sum of left and right channels at the listener’s ears. This causes audible comb filtering due to inter-channel phase differences, which the room reflections and reverberation are too weak to mask. (Playing in summed-to-mono is even worse than stereo, since in this case the comb filtering is introduced at the source, preventing any masking that would otherwise be provided by room reflections.) Reverb doesn’t help here, since inter-channel phase differences at the piano’s outputs will persist even after reverb is applied.

Finally, prioritizing high output and highly directional sound favors loudspeakers that use inherently resonant elements, e.g. an under-damped woofer cabinet and/or horn-loaded tweeter. These are efficient and plenty loud but the resulting spectral problems are hardly flattering to piano sounds, tending to make them harsh or strident.

Solutions

All of this suggests the need for a sound system designed along lines quite different from a typical keyboard amp / PA. The primary goal is to avoid audible comb filtering. This is most easily done by playing in mono, using only one of the keyboard’s outputs (and defeating the sum-to-mono behavior of the L-MONO jack by plugging a dummy cable into the unused output jack).

However, there are important reasons to prefer playing in stereo:

  • Neither the left nor right channel alone captures a good representation of the piano’s timbre. As I showed in another post both channels on my Yamaha CP50 suffer from spectral artefacts. (This isn’t surprising: a piano is a huge instrument; its near-field response at any point is bound to be corrupted by interference from reflections from the lid and body.) But by playing both channels, harmonic deficiencies in one channel can be compensated by the other.
  • Phase cancellation due to polyphonic interference typically occurs in only one channel at a time. Again, by playing both channels the spectral deficiencies in one will tend to be filled in by the other.
  • Reproducing both channels at different points in the room creates a more “complex” sound field, presenting somewhat different waveforms at each ear. This increases the listener’s perception of spaciousness and envelopment, which are highly prized elements of live performance [1].

(There is a common misconception that a piano’s “stereo” outputs create a stereo image on playback. They don’t. To my knowledge, piano samples are recorded with close-mic techniques inside the piano, so inter-channel phase and level differences don’t encode meaningful spatial information. It’s better to think of the two channels as providing two different snapshots of the piano’s tone; neither is quite right on its own, but they can be usefully combined to create a more nuanced picture.)

The downside of playing in stereo is that phase cancellation (comb filtering) is inevitable where the two channels sum acoustically at the listener’s ears. Fortunately this artefact can be masked by a sufficiently high level of reverberant field [2]. Most keyboard amps / PAs are unable to do this, since they aim to avoid exciting the reverberant field. Instead, for playing in stereo we need a source that projects a relatively low level of direct sound (though not so low as to obscure articulation and timing) and excites the reverberant field to a greater degree.

As a bonus, a sound system designed for increased reverberant level will not only mask comb filtering, it will also tend increase the level of early lateral reflections. This promotes the highly pleasant perceptions of spaciousness and envelopment associated with live acoustic performances [1].

Since both direct and reverberant fields will play a significant role in perceived timbre, the loudspeaker should be equalized in terms of both power response and direct response. This favors the use of constant-directivity loudspeakers, since these radiate the same spectrum in both the direct field and power response. Of these, the most common (and easiest to implement) are omni-directional (monopole) and dipole radiators.

Omni-directional Loudspeakers

An omni-directional loudspeaker aims to approximate a point source (monopole) which radiates the same intensity in all directions, at all frequencies. Compared with a more directional loudspeaker, this puts more acoustic energy into the reverberant field, raising its level by several dB. Small studio monitors (the ones with a 5″ or 6″ woofer) are often fairly omni-directional, which might explain why some players have had success with these.

However, the benefits of an omni loudspeaker are only available to listeners far enough away: a listener close to the loudspeakers (e.g. the player) will still experience a high level of direct sound. The following graph shows the variation of direct level with distance for several source types that radiate the same total power (assuming a critical distance of 1m, which is fairly typical but varies with room size and sound absorbing area):

The direct level falls off at 6dB per doubling of distance. For omni loudspeakers, if we want the direct level to be at least 6dB below the reverberant field (a bare minimum to mask comb filtering) then listeners need to be at least 2m away. This could work in a room with enough space, with a layout like this:

Ideally the direct-to-reverberant ratio should be adjustable, allowing for adaptation to the room acoustics and personal taste. This isn’t possible with an omni loudspeaker, but it does become possible with a dipole radiator.

Dipole Loudspeakers

To reduce the direct level below that of an omni source we need a source that directs sound primarily away from the listener (i.e. a loudspeaker with negative directivity). One way to implement this would be a pair of dipole loudspeakers (which radiate a figure-eight pattern) with the following layout. The dipole radiation pattern is shown in blue. The player is near the dipole null and so receives a low level of direct sound. The graph above indicates that at 75° off-axis, the direct sound is at least 6dB below the reverberant field even at distances less than 1m.

This setup has several advantages:

  • compact layout with loudspeakers near player; low level of direct sound provides monitoring for the player
  • relatively constant (and low) direct-to-reverberant ratio over a wide area that includes the player, masking comb filtering for all listeners
  • variable direct-to-reverberant ratio via orientation of the loudspeakers
  • high level of delayed lateral reflections (relative to direct sound) from side walls promotes sense of spaciousness and envelopment
  • proximity effect (inherent to dipoles) provides bass boost to the player only, which mimics the response at the player for a grand piano with the lid open

I’ve experimented with this setup using a pair of Linkwitz Lab LX521 dipole loudspeakers with my Yamaha CP50 stage piano. The results are excellent: for the first time I can play live in stereo without audible comb filtering, and the sound compares favorably to the experience in headphones. Also, as expected, even without added reverb there is a pleasant sense of spaciousness, envelopment and natural reverberation. Synthetic reverb is redundant.

Alas, dipole loudspeakers are a significant engineering challenge to design and execute, so cost becomes a significant factor (e.g. at the time of writing, components for a DIY version of the LX521 cost over $5000).

Conclusion

I’m convinced that a stereo pair of dipole loudspeakers is the best way to reproduce stereo piano samples in live performance. The downside of this approach is cost and complexity.

I’m working on a dipole implementation customized for this application, with the goal of making something portable and affordable.  Please let me know if this is something that interests you: sign up for my mailing list and I’ll contact you directly with updates.

References

[1] Floyd E. Toole, “Sound Reproduction: The Acoustics and Psychoacoustics of Loudspeakers and Rooms.” Focal Press, 2008.

[2] Søren Bech, “Timbral Aspects of Reproduced Sound in Small Rooms.” Journal of the Acoustical Society of America 97(3), p.1717-1726, 1995.

22 thoughts on “Toward High-Quality Live Sound for Digital Piano

  1. Richard, thanks for a well-thought-out post! I think, for the purposes of near player setup in a small room LXmini+ (LXmini plus LXsub) should be enough, and they can be built much cheaper than LX521.

    The toeing of dipole speakers on your picture looks unusual. Are they intentionally toed out? Where are the walls on the figure?

    • Hi Mikhail,

      Nice idea. I haven’t built the LXmini but did experiment with the Pluto and found that the near-player direct-sound level was too high, for reasons discussed above. I also built an experimental loudspeaker with a radiation pattern much like the LXmini — omni at low freq, transitioning to cardioid but with most of the energy directed away from the player. This was better in terms of raising the reverberant level. But the transition to omni at low freq created a muddy, bottom-heavy sound that I couldn’t fix with EQ. I wonder if the LXmini would lead to the same issue?

      Yes, the toe out in the picture is intentional. The conventional toe in raises the level of direct field in the audience — the opposite of what I’m trying to accomplish. The toe out also increases the relative level of reflections from side walls, which is a good thing in this application. Research quoted in Toole’s book suggests that listeners consistently prefer lateral reflections whose level is higher than the direct sound.

      • Richard, you are right about the radiation pattern of standalone LXminis, but since they start rolling off at 40-45 Hz, alone they can’t be used for reproducing full piano frequency range. What I was talking about is LXmini + Linkwitz subs. There are two models: one is LXsub4, same as in LX521, another is LXsub (single-driver version). Both models are dipoles. The resulting systems are called LXstudio and LXmini+2.

        • OK I follow you, but all the same I don’t think a sub is necessary. In theory a piano goes down to 28Hz but there’s actually very little energy below 60Hz. In another post I show measurements where I mic’d a 6′ grand and recorded some stuff using the full range of the keyboard. After applying a 60Hz low-pass filter (to derive the signal that would go to the sub) the waveform has peaks that are typically 40dB below the original. So it seems unlikely that I would miss having the sub. If I try really hard I can get the sub-60Hz signal up to -20dB, which might be worth reproducing. But I’ve never played the instrument that way in a musical context.

  2. Hello Richard,
    Thanks for yet another very interesting and useful post! Even though I don’t have a specific interest in electronic pianos, I find your insight into acoustical effects extremely valuable.
    I just wanted to suggest that other DIY options than SL’s LX521 exist, which could be a better fit for smaller wallets. An example is Visaton’s NoBox170, which can be built for around $500 a pair and driven with a simple stereo amp. It does have its shortcomings (notably the non-quite-symmetrical and therefore non-quite-dipolar structure), and certainly doesn’t have the same output capability as the LX521, but for domestic use it could be a very cost-effective option to the same effect.

    Let me also take this opportunity to belatedly thank you for all the bits and pieces you shared on using PC-based DSP to drive multi-way system. I found them invaluable in my own speaker prototyping endeavours.

    • Oliver,
      Thanks for the positive comments! I’ll look into the options you suggest. The not-quite-dipole pattern is likely a problem for me, as I want to put listeners pretty far off-axis. Even the very well-executed dipole designs have an irregular response anywhere near 90 degrees off-axis: the null is never perfect, and tends to move around with changing frequency. I’m working on a design aimed at better-behaved off-axis response; I’ll post about it here when I get the wrinkles worked out.

  3. Hi.
    Interested in your research. Have you tried Pianoteq in your experiments? Demo version of Pianoteq Standard allows you to change the configuration of microphones. Mono, stereo, etc. Could once again confirm their findings.
    And the piano recordings of the piano also suffer from these defects? How do sound recordings solve this problem?

    • I haven’t tried Pianoteq yet, though it’s been on my to-do list for a while. Until now I didn’t realize they have a free demo version; I’ll move this up the list!

      No, recordings of piano generally don’t suffer from these defects. For stereo recordings the mikes are more distant from (so roughly the same distance from) the source, so inter-channel phase differences don’t cause comb filtering. For close-miking, typically only one (mono) mic is used so again there’s no opportunity for comb filtering between mic signals; but there will always be some notes whose spectrum is somewhat compromised by local features of the sound feature, and I believe this is why larger mic diaphragms are preferred here. Sometimes a recording engineer will use several mikes in and around the piano (like Pianoteq) and in that case a naive mix can easily lead to comb filtering artifacts; great care is needed to come up with a good mix of the mic signals.

      Digital piano samples are recorded with close-miking to avoid picking up reverberant sound from the room. Doing this with two mikes is the worst-case scenario for creating comb filtering artefacts, and that’s unfortunately what we have with these instruments. However, I’m totally convinced a playback system can designed to mitigate this problem.

      • I’m going to build speakers for my piano with Pianoteq. I consider open baffle and omnidirectional. You are exploring ways to improve the quality of digital piano playback. I would be grateful for your thoughts and comments. I am now interested in:
        1. How to choose acoustic design (open baffle or omnidirectional)?
        2. How many lanes to choose?
        3. Useful or harmful high resolution detail acoustics. Is it possible to emphasize all the shortcomings?
        4. Will 8 “be enough for low frequencies?
        5. What are the most promising dynamics?
        etc.

        • Hi Amad,
          It really depends on how this will be used. Playing for yourself? or for an audience? small room or large? As to your questions:
          Q1. If for an audience in a larger room I would go with omni. Less demanding situations benefit a lot (in my experience) from using a dipole off-axis to reduce the level of direct sound relative to reverberant field.
          Q2. Maybe something got lost in translation… if by “lane” you mean “channel”, I would also go with 2 channels if possible.
          Q3-5. Ditto; I’m afraid I don’t understand your questions. Feel free to rephrase and I’ll do my best to answer!

  4. Hi.
    Interested in your research. Have you tried Pianoteq in your experiments? Demo version of Pianoteq Standard allows you to change the configuration of microphones. Mono, stereo, etc. Could once again confirm their findings.
    And the piano recordings of the piano also suffer from these defects? How do sound recordings solve this problem?

    • Hi Jasper –
      Sorry, somehow missed your comment before now. Better late than never…
      I haven’t tried Pianoteq or any of the other software pianos, but these are a possible solution. I think the ideal would be to (virtually) place close-spaced stereo mics a significant distance from the piano. Then phase issues are not likely to be a problem. Confirming this out is certainly on my to-do list.
      I don’t know much about how recording engineers deal with piano. I’ve recorded enough myself to know that it’s difficult to get right — mostly due to comb filtering issues thanks to the instrument being so large. Mono is easiest, but still has gotchas. If 2 mics are used then there needs to be a lot of attention to mic placement and post-processing to avoid the comb filtering issues that always crop if when combining 2 signals captured at different points. Putting the mics far from the piano sounds most natural, but this captures a lot of room sound which isn’t always desirable.
      I’ve heard lots of good piano recordings, but also plenty of not-so-great ones that suffer from mic placement issues.

  5. So what if you were to midi the keyboard through a DAW and have it come out an actual stereo system (not a PA). That seems to sound great, why is it different?

    • If the PA is stereo (usually not the case) then there’s no difference. But there’s a huge difference between a stereo system and a mono one, at least if you’re running stereo samples — mostly caused by comb filtering due to summing of L+R channels. Comb filtering can also mess up the sound even in stereo, but this seems to be hit-and-miss, mostly depending on the room.
      One possible difference is if your midi-to-DAW is running high quality samples or software piano; these are usually better than the on-board samples in most pianos. If they’re engineered with careful attention to mono compatibility then they’ll sound much better (whether in stereo or mono) than the built-in samples in a digi piano.

  6. The Yamaha DXR 10 powered speaker has a line input that accepts 2 TRS plugs
    ( as L and R from my Digital Piano)
    Does it make sense to plug 2 leads (L and R) into the same powered speaker to avoid comb filtering ? ( I only own one speaker) ?
    Thanks,
    Jay

    • Hi Jay –
      This is almost certainly not a good choice. My experience of mono amps having 2 inputs (some “keyboard” amps offer this) is that they simply sum the L and R channels. This is the worst-case scenario for causing comb filtering. It removes some information (spectral components) from the input signals, after which there’s no way to put it back.
      A far better option is to use only the L or R output alone — whichever sounds better. If you use the L output then you probably need to insert a dummy plug into the R output, to prevent your piano from summing L+R to the L/mono output.

  7. Hello,

    Very interesting! Have you think or already tested Distributed Mode Loudspeakers (DML). They are strange things with an uneven response, too short in the bass but with something unique (at least for me); piano music is just… natural. Like having a piano in the room. Hoping EQ can correct the major troubles. No doubt a sub can make the job in the bass

    • Hi Christian,
      I’m aware of DML speakers and have considered them as an option, but I haven’t actually tried them. I’m very interested to hear that your experience with them has been positive. Unfortunately EQ is not likely to help (their response is extremely resonant/modal). As for bass response, a sub isn’t need to cover the frequency range of a piano; a small woofer should do.
      Did you build your own DML or use a commercial product?

  8. Hi,
    Very happy to find your blog. I’m using a DP as midi controller with iOS piano apps (Korg Module, Ravenscroft, Pure Piano). The sound is excellent through headphones but has an annoying thin-ness through external speakers (inexpensive PC speakers with subwoofer that sound good for recorded music). Not sure if it’s exactly the phenomenon you’re talking about here.

    From the previous comment I discovered DML speakers and so made a pair this weekend using this method, but with different “exciters” and thicker foam (what was available): https://www.youtube.com/watch?v=gGzNkUmPdXc

    I don’t get the annoying thin-ness through the DMLs but the sound lacks some detail, clarity and general quality (my DMLs could undoubtedly be improved). Mixing the PC speakers sound with the DMLs gives better results; changing the location also changes the sound quite a bit. It all makes me wonder if simply mixing two sets of normal speakers and adjusting the postition and volume levels of each could lead to better sound.

    • Hi Zach – Thanks for sharing this… I’ve been meaning to run that experiment for a long time. Your results are what I was expecting/hoping for. This is encouraging!

      The annoying thin-ness you’re getting through conventional speakers is likely due to the phase issues I’ve written about; it shows up consistently across every combination of DP and loudspeakers I’ve tried. You’re the first to mention it in connection with software pianos. I kind of expected these to do better, but haven’t had time to check it myself.

      DMLs should help by decorrelating the L & R channels from each other and from the room reflections. This masks the phase issues. Alas, the clarity/quality issues with your DML aren’t surprising: DML aren’t great for transient response (important for piano, since it’s a percussive instrument) and they also lack output in the bass-to-lower-midrange. Maybe turn up your sub to compensate, and set its crossover frequency as high as you can?

      Your multiple-loudspeakers idea will probably cause more phase issues than it solves. Do let me know if you try it! But if the extra speakers are directional and they’re aimed *away* from the listener then this can sound really good with DP — I’m working on something similar now using dipole speakers. This does something similar to a DML, but uses the room reflections to decorrelate the channels. You can probably get something like this by listening to your DMLs “on edge” (where they don’t output much sound): this should increase the reverberant field relative to the direct sound, hiding the phase issues even more.

  9. Hi Richard,
    I found this article thanks to a recent mention in the Pianoteq forum. Thank you for all this work you have done on this problem!
    In your travels, have you been able to discern (or guess) what kind of tricks DP designers have had to use to get their built-in speaker systems to sound good? (In higher-end pianos, some of which sound pretty nice)
    I am in the process of trying to get Pianoteq to sound good on speakers in the loungeroom. With the info you have covered here I have some more leads to follow. Thank you again.
    Matt

    • Hi Matt,
      There’s a lot of variation between individual DPs, both in terms of how they are sampled (or modelled or synthesized) and in their on-board sound system. Some of them do sound much better than others. My assessment at this point is that the sound system matters much less than the quality of samples. The worst DPs have pretty bad phase issues in their samples, with lots of phase cancellation when they mix down to mono — this will sound sub-par through pretty much any sound system.
      The sound system requirements are quite modest, but I haven’t looked carefully at variations between DPs — my focus has been on stage pianos that use an external sound system, giving the player a chance to customize and optimize a bit.

Leave a Reply

Your email address will not be published. Required fields are marked *