The four tests
The tests follow the order and the kinds of question of the common Grade 1 to 3 aural syllabus, such as ABRSM's. Auraldesk is not affiliated with ABRSM, Trinity College London, AMEB, RCM or any exam board. Every phrase and passage is generated by this site for practice; no exam board's tests or books are copied.
- A. Clap the pulse. An 8-bar piece in 2 or 3 time plays. Then the screen counts one bar in at the same speed and the student claps the beat for 4 bars, a little louder on beat 1, while nothing plays. Then they tap 2 time or 3 time. In an exam the student claps along with the playing; here they clap straight after, because the microphone would otherwise hear the piano as well as the claps, and because an open microphone makes an iPhone's playing quieter. "Hear it again" lets them clap along, unmarked.
- B. Sing back. For each of 3 echoes the key chord and the first note play, then the 2-bar phrase twice. The screen counts one bar in and the student sings it back, on "la" or any syllable.
- C. Spot the change. A 2-bar phrase plays, then after a bar's rest the same phrase with one change. The student taps the bar that changed, and from Grade 3 whether the pitch or the rhythm changed. Marked exactly.
- D. Listen for a feature. A 4-bar piece with one clear feature plays, and the student answers one question about it. Marked exactly.
How the music is generated
Every test is made from a random seed by the rules below, so each attempt is new and can be played again exactly. Echoes are placed so every note lies from B 3 to E 5 (middle C is C 4), a range most children and higher voices can sing; lower voices sing an octave down, which counts. Speeds are multiples of 4 beats a minute. Echoes end on the keynote with a note of at least a beat.
| Grade | Echo keys (major) | Echo notes | Rhythms | Speeds (beats a minute) | Other keys used | Spot the change | Feature questions |
|---|---|---|---|---|---|---|---|
| Grade 1 | C major, G major, D major, F major, B flat major | the keynote to the third above (do, re, mi); moves of at most 1 scale step | crotchet, two quavers, minim | echoes 80 to 96; clap the pulse and features 84 to 112; spot the change 80 to 96 | none | a pitch change | loud then quiet, or quiet then loud (about 10 dB apart); smooth (joined) or detached (short and separate) |
| Grade 2 | C major, G major, D major, F major, B flat major, A major, E flat major | the keynote to the fifth above (do to so); moves of at most 2 scale steps | crotchet, two quavers, minim, dotted crotchet and quaver | echoes 80 to 104; clap the pulse and features 84 to 120; spot the change 80 to 100 | A minor, E minor, D minor | a pitch or a rhythm change | loud then quiet, or quiet then loud (about 10 dB apart); smooth (joined) or detached (short and separate); getting louder, getting quieter, or staying the same |
| Grade 3 | C major, G major, D major, F major, B flat major, A major, E flat major, E major, A flat major | the fifth below the keynote to the fifth above (low so to so, an octave); moves of at most 2 scale steps | crotchet, two quavers, minim, dotted crotchet and quaver, two quavers and a crotchet | echoes 84 to 112; clap the pulse and features 72 to 132; spot the change 84 to 104 | A minor, E minor, D minor, B minor, G minor, C minor | a pitch or a rhythm change, and which kind it was | loud then quiet, or quiet then loud (about 10 dB apart); smooth (joined) or detached (short and separate); getting louder, getting quieter, or staying the same; getting slower, getting faster, or keeping the same speed (from halfway) |
Clap-the-pulse pieces stress beat 1 of every bar in the tune and the bass. A pitch change moves one note, never the last, up or down one scale step; a rhythm change swaps the lengths of two neighbouring notes in one bar, keeping both notes: two crotchets become a dotted crotchet and a quaver, or back, or a minim and a crotchet change places (this can move or shorten the last note, which keeps its pitch). Each bar, and at Grades 2 and 3 each kind of change, is the answer equally often. A feature piece changes only the feature its answer names; everything else stays at a steady speed, level and touch. Grades 4 and 5 are being prepared and will be added after review.
How singing is marked
- The count-in bar gives the background level. A take whose singing is not at least 12 dB above it is not marked: the page says so and suggests a quieter place, and the student can mark it themselves.
- The pitch is tracked every 10 ms with the YIN method (70 to 1,100 Hz, threshold 0.15), written in TypeScript and tested against synthesised tones.
- The pitch line is cut into notes where the sound dips (the "l" of "la") or the pitch moves more than 80 cents and stays there for 60 ms. Two touching pieces (no silent gap and no dip between them) are one note when the joined pitch stays within 80 cents of its middle for nine tenths of the time and swings across the middle at least 3 times: strong vibrato, not a step to the next note. Pieces shorter than 80 ms join the note they slide into or are left out. A note that follows straight on from the one before starts where its new sound comes in, which in an echoing room is before its pitch is seen to change. A note's pitch is the middle of its steady part.
- Sung notes are paired with the played notes by their order, allowing a missed note or an extra one, weighing pitch and, more lightly, time. So a right note sung late is still paired with its own played note: it is marked in tune and late. A pair more than 400 ms from its place is kept only when the note before it was held on into it, or a neighbouring pair is shifted the same way; otherwise it is not paired.
- One timing offset is set for the whole echo, used only for the rhythm mark: the median of the sung start minus the played start over the paired notes (the lower middle value when there is an even number). The playing's speed is kept; nothing is re-fitted note by note.
- Each played note is then marked twice. Pitch: right when its sung note is within 50 cents of it in any octave (a cent is a hundredth of a semitone), so singing an octave down counts. Rhythm: right when its sung note starts within 100 ms of the played start plus the offset. A missed note is wrong in both.
- A sung note paired with nothing is an extra note. It counts against the pitch mark when it lasts at least 120 ms, is no more than 10 dB quieter than the sung notes, and is not within a semitone of the note before or after it (that is the room's echo or a note sung twice).
- The two marks are shown apart: pitch is notes in tune out of the notes played plus the extra notes; rhythm is notes on time out of the notes played. So right notes in the wrong rhythm get credit for the notes. The score is the two, half each, over the 3 echoes.
How clapping is marked
- The count-in bar gives the background level; claps less than 12 dB above it are not marked.
- Claps are found by spectral flux (how suddenly the sound above 500 Hz gets louder) with a threshold set by the sensitivity slider, at least 100 ms apart, each timed to the sample where it starts. Other sudden sounds in the room (a cough, a knock, speech from a television) can be heard as claps; they show as extra claps on the timeline.
- The beat grid comes from the music's speed and the count-in. One offset is set for the whole take: the median distance of the claps from their nearest beats, used only up to 150 ms either way. That allowance is for the delay of the speaker, the microphone and the eye on the count-in; a set-up with a longer delay (some Bluetooth headsets) has its claps marked late.
- Each beat takes at most one clap: within 70 ms of its beat it is on the beat, otherwise early or late; a second clap for the same beat is extra. Joining late in the first bar is fine. The pulse mark is claps on the beat divided by the beats to clap or the claps made, whichever is more.
- The strong beat shows when, in at least 60 percent of the bars, the beat 1 clap is at least 2 dB louder than the others.
- The score is the pulse mark for half, the strong beat for a quarter and the 2 or 3 answer for a quarter.
Spot the change and features are tapped answers: the bar for the whole mark (half each for the bar and the kind at Grade 3), the feature right or not. In the mock test each tapped answer has 20 seconds. The weak-area list is every area averaging under 70 over the last 30 days, lowest first.
How well the marking agrees
The marking was checked against clips of generated singing and clapping, each built to deserve a known mark, played into the check page through a browser's microphone input. Two sets were made by independent reviewers with their own generators and labels; the others by this site's own generator. No real voices, real rooms or real phones were used: no recordings of people singing or clapping, and the marking has not been checked on a real iPhone or Android phone. A clip agreed when every note's pitch and rhythm mark, every clap's mark and the strong beat all matched. Over the two independent sets together, 70 of 75 clips agreed (66 of 75 against their labels as first written), and no clip built to be wrong got the full mark:
- A first independent set, made by a reviewer with a different voice and clap generator and their own labels (39 clips): 36 of 39 clips agreed (92.3 percent). 35 of 39 against the label file as written; one label was later agreed to be a slip (a note sung in tune and on time). The 3 that differed: one note timed 160 ms late in a very echoing room, speech from a television heard as extra claps, and a strong beat not found on claps of varying sharpness.
- A second independent set, made by a second reviewer with a third generator, including held notes, swapped notes, coughs, a phone speaker and strong vibrato (36 clips): 34 of 36 clips agreed (94.4 percent). 31 of 36 against the labels as written; 3 clips of claps that speed up were labelled clap by clap, while the marker counts a clap that lands a beat early as on that beat, which the reviewer accepted. The 2 that differed: a short note with strong vibrato marked 140 ms late, and a cough counted as 2 extra claps.
- A set made by this site's own generator after the first review, with rhythm errors of 110 to 250 ms and claps 80 to 120 ms off the beat (30 clips): 29 of 30 clips agreed (96.7 percent). The one that differed was a repeated note sung joined, without a new start, heard as one note.
- The first two sets, made by this site's own generator with its own labels before the first review (72 clips): 71 of 72 clips agreed (98.6 percent). The one that differed was an echo sung 6 percent faster than the playing, whose last note came more than 100 ms early.
Rooms, voices and devices vary more than generated clips do, which is why every mark shows what it heard and can be marked by the student instead.
Seeing what the marker heard
Every mark shows its evidence: the sung pitch as a line over the played notes, each drawn as a band 50 cents either side, with a table of every note; the claps on the beat grid, with a table of every clap. You can hear your take again. If a mark looks wrong, press "The mark was wrong" and mark it yourself: your mark is the one saved, and the microphone's is kept beside it in the log, in this browser only.
The microphone, the speaker and headphones
- The browser asks for the microphone only after you press a button to start a test or a check.
- For singing, the browser's echo cancelling, noise reduction and automatic level are turned off, so the pitch is heard as sung. For clapping they are off too: they soften the sharp start of a clap.
- The microphone is closed whenever the piano plays: it opens only after the last note's sound has fully stopped, release included, plus 50 ms and the speaker's own delay as the browser reports it (or 100 ms when it does not). On iPhones an open microphone makes the speaker quieter and rougher (a known Safari problem), and an open microphone would also hear the piano.
- The count-in bar clicks softly as well as showing the beats. Those clicks are the one sound made while the microphone is open; they all end before you start, and the marking never looks at that bar.
- The 4 second speaker check on the check page plays 3 tones while the microphone listens. Heard (10 dB or more above the quiet before): the set-up works. Not heard: headphones are on (fine, and often better for marking) or the sound is going somewhere else, such as a Bluetooth speaker in another room.
- Bluetooth headsets with their own microphone can switch to a low-quality call mode and add a delay. Wired headphones, or the device's own speaker and microphone, mark best.
- Nothing is recorded to keep: a take stays in the page's memory for its result and is gone after the next take or when the page closes.
What it is not
- The marker hears pitch and timing only. It cannot judge tone, expression or musical sense; a teacher can.
- It cannot tell a cough or a sound in the room from a clap, or a television from the background: those can cost marks, and they show on the timeline.
- It is practice, not an exam result or a prediction of one. Exam marking is done by examiners against the exam board's own criteria.
- It does not use speech recognition: answers are tapped.
- It does not copy any exam board's tests or books; everything is generated here.
- It is not a teacher. The teacher sheet is there so a teacher can see what was practised.
What is free
Free: Grades 1 to 3, 2 attempts of each test a day (from local midnight), the check page, the pitch trace and clap views, marking yourself and 1 student. Premium adds unlimited attempts, the mock test, the teacher sheet, up to 5 students and Grades 4 and 5 when they are added.