I sang the exact note but scored 65. Why?
Absolute pitch is not scored. Mimic Party scoring transposes your voice into the reference's range and compares the contour shape. Your Timing or Attack bar was likely the low one.
From microphone to score in five steps
Mimic Party scoring runs a five-stage pipeline on every take. First, the browser captures your microphone at 16 kHz mono. That is enough detail for voice shapes, and light enough for phones. Second, the scorer extracts four features at 20 frames per second: an RMS envelope (your volume curve), a pitch contour (where your voice rises and falls), onset positions (where each hit lands), and a seven-band spectral shape (how bright or dark your tone is). Third, it trims the dead air before and after you speak, so reaction time and trailing breath do not count. Fourth, it stretches your take onto the reference's time axis using dynamic time warping, which means you do not need to hit the exact tempo. Fifth, it computes three weighted distances and produces the score.
The four scoring profiles
Every sound in Mimic Party belongs to one of four profiles. The profile decides which three dimensions get scored and how much each one counts. You can see the profile on the cue card, but the weights stay hidden.
Pitch: melody and movement (45 / 35 / 20)
Pitch sounds weigh the contour at 45%, timing at 35%, and attack at 20%. The scorer transposes your voiced frames into the reference's register before comparing the shape, so a deep voice can copy a high clip. The transposition itself costs about 16 points per octave on the contour bar. Unvoiced frames stay at zero.
Rhythm: beats and spacing (45 / 30 / 25)
Rhythm sounds weigh onset count and placement at 45%, timing at 30%, and energy at 25%. The scorer detects spectral-flux peaks to find where each hit lands, then compares the count and positions against the reference. Missing a beat hurts more than being slightly off-tempo.
Noise: texture and brightness (40 / 35 / 25)
Noise sounds weigh spectral shape at 40%, timing at 35%, and attack at 25%. The seven-band energy ratio captures whether your tone is bright or dark, regardless of volume. Two takes at the same volume can score differently if one is breathy and the other is guttural.
Speechlike: syllable shape (40 / 35 / 25)
Speechlike sounds weigh contour at 40%, timing at 35%, and duration at 25%. A 35% duration tolerance either way means a one-second clip accepts anything from 0.65 to 1.35 seconds, and short clips get a 600 ms floor so a 300 ms bark is not judged on a single frame.
What each score bar measures
After each take, the verdict shows three bars. A low bar is more useful than the total score because it points at the one thing to change next time.
| Bar | What it compares | If it is low |
|---|---|---|
| Pitch / Contour | The rise and fall of your voice, after transposition | Decide up, down, or both before you start |
| Timing | 55% duration match plus 45% onset rhythm | Match the clip length in one breath |
| Attack | The first quarter of your aligned envelope | Hit the first syllable at full volume |
| Rhythm | How many hits and where they land | Count the beats while the clip plays |
| Energy | The loud-and-soft pattern across the whole take | Follow the reference dynamics, not one flat volume |
| Shape | Seven-band brightness compared band by band | Change how you make the sound: breathy, rolled, popped |
| Duration | Active sounding time, 35% tolerance, 600 ms floor | Stop where the clip would have stopped |
Why some takes get no score
Before any comparison, Mimic Party scoring checks whether your recording is clean enough to judge fairly. If it is not, the game refuses to guess. You keep your life and streak, and the same sound comes back for a free retry. Four reasons block a score: too quiet (the whole take stayed under -45 dB), clipped (more than 1% of samples hit the ceiling), noisy (background at -35 dB or louder with almost no voiced sound), and too short (less than 80 ms of actual sound). Each has a clear fix shown on the verdict screen.
How the final score is calculated
The three bar scores are not a simple average. Each bar starts as a distance from 0 to 1, where 0 means identical and 1 means completely different. The scorer flips that into a similarity percentage and applies a score floor of 50, so only the top half of similarity earns points. A distance of 0.5 or worse scores zero on that bar.
The final number is the weighted sum of the three bars. For a Pitch take scoring 76 on Pitch, 84 on Timing, and 70 on Attack, the math is 76 × 0.45 + 84 × 0.35 + 70 × 0.20 = 77.6, rounded to 78.
Score tiers and party points
The verdict screen labels your take with a headline. NAILED IT starts at 90, ROOM SHAKEN at 75, SOLID COPY at 60, and anything below is GLORIOUS CHAOS. The clear line is separate: Today's Stage clears at 30, and Survival clears at the current depth target, which climbs from 30 to 70 as the run goes deeper.
Party points are what go to the leaderboard, and they have their own formula: score × route multiplier × wheel multiplier + streak bonus. Safe route pays 1.0×, Risk pays 1.5×, and every third consecutive clear adds 10. An 80 on Risk as your third clear pays 80 × 1.5 + 10 = 130 party points.
How to read the waveform comparison
The verdict screen shows two waveform lanes on a shared time axis. The top lane is the reference envelope; the bottom is your aligned take. If the left edges do not match, Attack is low. If the shape looks right but the lanes differ in length, Timing is the problem. If the rises and falls do not line up, Energy is off. Use the waveform to find the gap, then check the bars to confirm.
Strategy tips by profile
For Pitch sounds, trace the contour in your head before you start. The direction matters more than the exact note. For Rhythm sounds, count the hits with your fingers while the clip plays, because onset count carries 45% of the weight. For Noise sounds, change the texture (breathy, guttural, popped) rather than the volume, since the seven-band ratio measures tone. For Speechlike sounds, finish in one breath and stop where the clip would have stopped, because trailing breath eats into your Duration bar.
Frequently asked questions
Does a higher or lower voice get a lower score?
No. Transposition moves your voiced frames to the reference register before comparison. One octave off costs about 16 points on the contour bar, no more.
Does the scorer recognize my voice?
No. Mimic Party scoring extracts four numerical features: envelope, pitch contour, onset positions, and band energies. It never identifies who you are, and audio is released after scoring.
That is the full picture behind Mimic Party scoring. If you remember one thing, make it this: timing sits between 30 and 35% on every profile, so getting the length right beats any detail of how the sound is made. Play Mimic Party Online now, or read the full How to Play guide.
Last updated 2026-09-09. Written by the Mimic Party team.