METHODOLOGY · UPDATED SEPTEMBER 2026
How Meme Party voice scoring works
The score is produced from audio features, not from the words you say or the identity of your voice.
The five signals
Before a game starts, Meme Party extracts features from each reference sound. It runs the same extraction pipeline on a submitted take, aligns normalized time series, and converts their similarity into component scores.
•
Melody: the shape of detected pitch over time, normalized around the median pitch.
•
Rhythm: the timing and pattern of attacks or onsets.
•
Energy: how loudness rises and falls through the take.
•
Duration: how close the recording length is to the reference.
•
Tone: the median fundamental pitch, folded by octaves to avoid over-penalizing different voices.
Adaptive weighting
For a strongly tonal reference, melody can contribute up to 36% and tone up to 12%. Duration always contributes 10%. The remaining weight is split 55% toward rhythm and 45% toward energy.
When a reference has little reliable pitch, melody and tone receive less weight and the system redistributes that importance to rhythm and energy. This prevents a noise, impact, or spoken sound from being judged like a sung note.
Silence and score range
A recording shorter than 0.05 seconds is scored zero. A take that is both too quiet and insufficiently active is also scored zero. Valid component similarities are combined and shaped into an integer total from 0 to 100.
Scoring is deterministic for the same decoded audio and reference, but microphones, room noise, clipping, echo, and browser encoding can affect the extracted signal.
What the score does not do
The scorer does not transcribe speech, identify the player, infer emotion, or decide whether an imitation is funny. It measures similarity. The playback and the people in the room decide the rest.
FREQUENTLY ASKED QUESTIONS
Do deeper voices score worse?
Not by design. Melody is centered around each recording’s median pitch, and the tone comparison is octave-folded to reduce penalties caused only by vocal range.
Why can background noise change my score?
Noise can alter detected attacks, energy, and pitch confidence. Headphones and a quiet room produce a cleaner comparison.
Is the score generated with speech recognition?
No. The scorer analyzes acoustic features and does not convert the recording into words.
Published by Meme Party. Updated September 16, 2026.
Meme Party is an online voice imitation game to play with friends.
© 2026 Meme Party