You load three upright piano libraries, play a few chords, and the one that is a couple of dB louder wins. That is not a verdict on the piano. It is a verdict on your monitor level.
Readers keep asking which library to pick. A forum thread titled "Best upright piano library" was started on Audiosex on November 19, 2026 (https://audiosex.pro/threads/best-upright-piano-library.85920/), and it is one more place the question comes up. This article does not use any replies from it, and it does not rank anything. Demo videos and recommendations conflict because every demo uses different reverb, mics, and loudness. The fix is a test you control, and this one works for 2 to 4 libraries.
What you need before you start
You need a DAW, the libraries you want to audition, a loudness meter that reads LUFS, and decent headphones or monitors. Demos work if they play back the same sounds you would buy, but check for time limits and muted notes. Step 8 needs a vocal or a rough mix of your own. A friend helps with the blind step but is not required.
Phase 1: Build the test template
Step 1: Write one fixed MIDI clip
Create a single MIDI clip of about 60 to 90 seconds with four sections. Leave a bar or two of silence between them so release tails can finish. The velocity numbers below are starting points, not rules. Adjust them to your keyboard, but never change them between libraries.
- Soft passage: slow single notes and a few quiet chords, velocities roughly 20 to 40, spanning low, middle, and high registers.
- Mid-velocity passage: a simple melody with chords, velocities roughly 60 to 85, the kind of part you would actually write.
- Repeated-note passage: the same note at least 8 to 16 times in quick succession, at one fixed velocity, then again at a second velocity. Repeat on a low, a middle, and a high note.
- Pedal passage: chords with sustain (CC64) held across several changes, then released cleanly, then a few half-pedal moves if the library supports them.
Put a slow scale from soft to loud somewhere in the clip. It exposes velocity-layer jumps faster than any melody does.
Copy the clip onto every track instead of re-recording it. Quantized, edited MIDI removes human variation from the comparison.
Step 2: Strip every library to a neutral state
Turn off all built-in reverb, EQ, tape, saturation, compression, and room or convolution effects in each plugin. Disable any stereo widening. Leave each library's velocity curve at its default, because that curve is part of what you are judging.
Some libraries bake room sound into the samples. You cannot remove that, so note it on your scoring sheet instead of pretending the test is dry. This is an inference from how multisampled instruments are commonly built, so check each library's manual for what its mic and room controls actually do.
Step 3: Match mic perspective
Pick the closest equivalent in each library. If one offers close, player, and room mics, start every library at the same perspective, for example a close-plus-player blend with the room mic down. If a library has only one fixed perspective, keep it and write it down.
Exact matching is rarely possible. The goal is a fair starting point. Record each library's mic settings next to its name and leave them alone for the rest of the test.
Step 4: Route everything the same way
Send each library to its own stereo audio track, all feeding the same bus or going straight to the same output. Put no processing on the master. Keep the same sample rate and buffer size throughout. This pairs well with see also: how to test audio to midi converters on your own tracks.
Leave the plugin's internal output volume alone until step 5. For level matching, use a clip gain or trim stage placed after the plugin.
Phase 2: Level-match before you listen
Step 5: Match loudness to within about 0.5 dB
Loudness bias is the biggest trap in this whole exercise. A louder version tends to sound fuller and clearer, so an unmatched A/B tells you which library is louder, not which is better.
Solo one library at a time and play the full test clip with a loudness meter on the output. Note the integrated LUFS reading. Pick one library as the reference, then trim the others until each lands within about 0.5 dB of it. Measure again after adjusting, since one pass is rarely exact.
Watch peak levels too. If one library needs a big trim to match, its dynamics are probably more compressed or more extreme than the others. Write that down as a finding, then continue.
A meter reading is a good start, but you can still verify by ear. Switch quickly between two tracks on a single sustained mid-velocity chord. If one still feels louder, shave the difference and re-measure.
Step 6: Bounce each library to audio
Render each library's output (after the trim) as an audio file of the full clip. This locks the settings, so a later plugin tweak can't change the test. It also lets you rename and shuffle files without seeing plugin windows.
Phase 3: Blind listening
Step 7: Hide names and randomize order
Rename the bounced files to neutral labels such as A, B, C, and D. Have a friend do this, or use a random list generator and keep the key in a text file you don't open until scoring is finished. Shuffle the track order, then shuffle it again for the second pass. This pairs well with our guide to varispeed vs time stretch for sampling: a daw exercise.
Listen to one section at a time across all libraries before moving on. Do the soft passage for every library, then mid-velocity, and so on. Hearing the same section back to back makes differences much easier to catch.
Take a break of at least 20 to 30 minutes. Then re-listen in a new random order and re-score without looking at your first sheet. If a score changes by two or more points, that item needs a third listen.
What to listen for
Note-to-note consistency: play the slow scale. Listen for single notes that sound brighter, duller, or louder than their neighbors.
Release tails: listen to the silence after each soft chord. Check whether tails decay naturally or cut off abruptly, and whether the tone changes as the tail fades.
Pedal and mechanical noise: in the pedal passage, listen for damper thumps, pedal creaks, and key noise. Upright pianos have characteristic mechanical sounds, and some libraries let you adjust them. Decide whether the noise feels like a real instrument or like a loop sitting under the notes. Either can suit different music, so score what fits your use.
Velocity-layer transitions: during the soft-to-loud scale, listen for audible steps where the tone jumps instead of rising smoothly.
Repeated-note machine-gunning: in the repeated-note passage, listen for identical samples firing in a row. If the same note sounds copy-pasted, the library may have too few round-robin variations at that velocity. This is an inference from how sample rotation works, so confirm it in the documentation if you want to be sure.
Also read: suno bridge with ableton live 12: try the beta or wait?
Phase 4: Test in a mix
Step 8: Put the survivors under a vocal or in your track
Solo piano flatters libraries with big low mids and generous room sound. Those same qualities can clog a mix. Drop each bounced piano under a vocal or into a rough mix of your own, keeping the level-matched trim in place.
Listen for three things. Does the piano stay out of the vocal's way? Does the left hand turn muddy? Does the top end disappear or get harsh? Add only the processing you would normally use, and apply the identical chain to each piano. If one library needs far more EQ than the rest to sit properly, that effort counts against it.
Run this step once on a quiet ballad and once on something busier. A piano can behave differently in a sparse arrangement than in a dense one.
The scoring sheet
Score each library from 1 (poor) to 5 (excellent) for every row, in both listening passes. Use the notes column for what you actually heard.
| Category | What a 5 sounds like | Pass 1 | Pass 2 | Notes | |---|---|---|---|---| | Consistency | Even tone and level across the keyboard | | | | | Noise character | Pedal and key noise feels believable and usable | | | | | Decay and release | Natural tails, no abrupt cutoffs | | | | | Dynamics | Smooth velocity response, no obvious layer steps | | | | | Repeated notes | No machine-gun effect | | | | | Mix fit | Sits under a vocal with minimal fixing | | | |
Average the two passes, but read the notes before you trust the numbers. A library with one 1 in a category you care about may lose to a library with fewer high scores. Decide the weights beforehand. If you write intimate, quiet pieces, soft-passage consistency and noise character matter more than loud dynamics.
Unblind only after scoring is finished. Then compare the top scorer with what you expected. If the library you hoped would win scored lower, that is exactly the bias this protocol exists to catch.
What to expect next
Your first run will probably cut the field quickly, since weak spots like scale jumps and machine-gunning show up fast. A second round with only your top two, using your own music instead of the test clip, should settle the close calls. Keep the template file. It works the next time a new library appears and someone asks which one to buy.



