Sign up free

Music performance assessment: recorded singing and playing

Set a recorded singing or playing task, mark it on a four-line rubric, hear twenty-eight takes in a lunch hour and give feedback a student can act on.

The EdProcto editorial teamWritten and checked against the research it cites. Last checked . How the guides are made

Guides · Question shapes and when to use them · · 13 min read

How do I assess a recorded music performance, sung or played, for a whole class?

A recorded music performance assessment works when every student records the same short extract, under a minute, at a stated tempo, and the teacher marks each take once on a rubric with separate lines for pitch, rhythm and pulse, tone and control, and expression. One marker's score is rough, so keep the stakes low and give one change to practise.

A music performance assessment by recording asks each student to sing or play one short extract into a phone and hand the take in. The teacher hears every student, which a lesson of twenty-eight rarely allows.

This guide is published by EdProcto, whose Sing or play shape does the job: the student records sound in the browser, and a class teacher or an instrumental tutor marks the take against a rubric. Talks are covered in recorded presentations, speech in spoken answers.

Every source was read on 10 October 2026: Tsay (2013), Beh et al. (2025) and the three web pages as full text, the rest as abstracts, with the method of Awan et al. (2025). Most study auditions, juries and graded exams, not class tasks.

What does research say about music performance assessment?

It says that judges separate pitch and rhythm more surely than tone and technique, that one marker's number is rough, and that the eye sways the ear.

Which criteria can a marker tell apart?

Intonation and rhythm, at least. Zdzinski and Barnes (2002) had 50 judges rate 100 recorded string performances by school-age players. Five factors came out: interpretation, intonation, rhythm with tempo, vibrato, and articulation with tone as a single factor. In Thompson and Williamon (2003), three experienced evaluators marked conservatoire students on a scheme modelled on a graded exam board's: its items rose and fell together, so the categories discriminated little.

Is one overall mark as good as marks by criterion?

Nearly, for the number. In Mills (1991), a segmented scheme accounted for 71% of the variance in the holistic marks that 29 assessors gave 10 performances. The case for criteria is feedback: a line tells a student what to practise, and a single mark cannot.

How well do examiners agree?

Too loosely for one marker to rank a class finely. Bergee (2007) recorded 8 high school wind players three times each, and 10 certified adjudicators scored the takes out of 100. Raters were a strong source of error, and five were needed to reach a generalisability coefficient of .80. Which take was heard, and in what order, made virtually no difference.

Does seeing the performer change the mark?

Yes. Across 15 studies, Platz and Kopiez (2012) found performances rated about half a standard deviation higher (d = 0.51) when seen and heard than when only heard. In Tsay (2013), professional musicians picking the winner among three competition finalists from six-second clips managed 25.7% from sound alone, below chance (33%), and 47.0% from silent video. A sound recording keeps the mark on what is heard.

Does listening back help a student improve?

It changes what students hear; alone, it has not been shown to change how they play. The 112 middle school players in Silveira and Gavin (2016) rated their pitch and rhythm differently once they had heard the recording. For the 82 junior high players in Hewitt (2001), listening to their own recordings had no significant effect, while those who evaluated themselves against a model recording improved more in tone, melodic and rhythmic accuracy and interpretation. In a five-week study of 211 pupils, teaching self-assessment did little for their playing (Hewitt, 2011).

Does recording ease performance anxiety?

Somewhat, in the one comparison we read, and with a pressure of its own. Beh et al. (2025) followed 28 music undergraduates whose first-year exams were recorded and second-year exams live. On an inventory scored out of 240, anxiety averaged 127 for the recorded exam and 139 for the live one. Most had recorded take after take in search of a perfect one, and one spent three days on a two-minute video.

How good is phone recording quality for music?

Good enough for the pattern of a performance and less sure for its tone, on the nearest evidence, which concerns speaking voices. Awan et al. (2025) compared four smartphones with a precision sound level meter on 24 speakers. The phones tracked the reference closely (correlations above 0.90), yet the balance of low against high frequencies differed strongly by device, and the room affected almost every measure. Their recordings used no automatic gain control or noise reduction.

A browser may add both. MDN (2026) describes three kinds of processing for a microphone: automatic gain control, which evens out volume; noise suppression, which filters background noise; and echo cancellation, which removes the device's own speaker sound. On music they may flatten a crescendo, thin a held note or swallow a backing track.

Trinity College London (n.d.), which examines grades from recordings, tells candidates to turn off automatic noise reduction, play an accompaniment through an external speaker and check the balance in a quiet room. So record the real instrument, loud and soft, on a phone like your students', and listen for what survives.

How do you set a recorded singing or playing task?

Fix the extract, the tempo, the takes and the place of the phone, so that twenty-eight takes differ only in the performing.

  1. Choose a short extract. Sixteen bars or fewer, under a minute.
  2. State the tempo. "Crotchet = 96, no slower than 84."
  3. Choose a click or an accompaniment. Four clicks fix the start and leave the pulse to the student; a track throughout holds it for them. Either plays from a second device, and the phone that records stays on the quiz.
  4. Settle the takes. Up to three, the last one sent being the one marked. Bergee's (2007) takes scored alike, and Beh et al. (2025) show where unlimited takes lead.
  5. Place the phone. One to two metres away, level with the instrument or the mouth, on something still; further back for brass and drums. Working rules, not findings.
  6. Say what is marked and who listens.
  7. Build it and record a take yourself. In EdProcto, choose Sing or play, type the task into "Prompt" and press "Preview as student".

What must families be told about the recording?

Who hears it, what it is for and when it is deleted. A child's voice on a recording is personal data. The Information Commissioner's Office (n.d.) in the UK says such data must not be held longer than needed, asks for a policy with standard retention periods, and notes that the UK GDPR sets no specific time limits. Your school's own policy decides, and nothing here is legal advice.

Published music and commercial backing tracks have owners too: check the terms before putting a score or a track at a public address.

What does a rubric for a recorded performance look like?

Four lines of four described levels, each something a listener can hear.

Line 4 3 2 1
Pitch Every note right and in tune One or two notes wrong or out of tune Several wrong notes, or tuning that drifts The tune cannot be recognised
Rhythm and pulse Steady at the set tempo; every rhythm right Steady; one rhythm wrong or one hesitation Pulse speeds or slows, or several rhythms wrong Stops and restarts
Tone and control Clear, even sound; notes start cleanly and last their length Mostly clear; a few notes crack, clip or fade Thin or forced for much of the take The sound breaks down
Expression Dynamics and phrasing as marked Some dynamics; phrases mostly joined One level throughout; phrases chopped Notes only

Each line scores 4 to 1, sixteen points in all, and no recording scores 0. Tone and articulation share a line because the judges in Zdzinski and Barnes (2002) rated them as one factor. Posture and hand position have no line: what cannot be heard belongs to the lesson. If your test take evened out loud and soft, judge expression by phrasing.

How do you mark twenty-eight recordings in a lunch hour?

Keep the extract under 45 seconds and hear each take once: two minutes a student, 56 minutes for twenty-eight. The minutes are our assumptions.

Step Sum Minutes
Listening once 28 × 45 seconds 21
Scoring four lines, one note 28 × 60 seconds 28
Opening and saving in EdProcto 28 × 15 seconds 7
Total 56

Every extra 15 seconds of music adds seven minutes, and a two-minute piece is 56 minutes of listening before a mark is written.

Pencil pitch and rhythm while the take plays, tone and expression at the end. Mark five again tomorrow, unseen: a line that moves two levels needs clearer wording. One marker's number is rough (Bergee, 2007), so sixteen points say what to practise and do not rank a class finely.

What feedback on a performance can a student act on?

A level for each line, one bar and one thing to practise.

  • Line scores first. "Pitch 4, rhythm 2, tone 3, expression 3."
  • One place, by bar number. "Bars 10 and 11: the quavers rush."
  • One thing to practise, with a tempo. "Clap bars 9 to 12 with a click at 72, then play them."
  • Their rating beside yours, made after listening back against a model recording, as in Hewitt (2001).
  • A second take within a fortnight.

A worked example: one extract, two students

This example is made up: the class, the task and the two students.

Sing or play bars 1 to 16 of the Ode to Joy theme from your part. Crotchet = 96, no slower than 84. Four clicks, then begin. Up to three takes: the last one you send is marked. Phone two metres away. Marked for pitch, rhythm and pulse, tone and control, and expression. Only your teacher listens.

Sixteen bars of four beats at 96 last 40 seconds. Two places sort the class: the pairs of quavers in bars 10 and 11, and the last note of bar 12, which Beethoven ties over the bar line so that the opening returns a beat early.

Line Student A Student B
Pitch 4: every note right and in tune 3: two flat notes in bar 12
Rhythm and pulse 4: steady at 96, the tie held 2: quavers rush; no tie
Tone and control 2: thin, long notes fade early 4: full, even sound
Expression 2: one level throughout 4: phrases shaped, the return louder
Total 12 of 16 13 of 16

B's take is the one a listener enjoys, and one impression mark would put B well ahead. Four lines show that A has the notes and the pulse, B the sound, and each gets different work.

The note for A: "Pitch 4, rhythm 4, tone 2, expression 2. Every note is right, the tie too. The long notes in bars 4 and 8 fade: hold each for its two full beats, then record bars 1 to 8 again." B's note names bars 10 to 13 and a click at 72.

How does EdProcto build and mark a recorded performance?

EdProcto is the best tool for a recorded performance when a lesson is too short to hear everyone, because each take arrives as sound alone, beside your rubric, with one box for the points and one for the note: marks on what is heard, feedback by criterion.

  • Building. Sing or play is in the builder now. "Prompt" takes the task and "Rubric (what earns the points)" the four lines; set "Points" to 16. For a model recording or the score on the same screen, "Another shape" adds Listen and answer, which has "Audio clip address", or Caption it, which has "Image address". Switch that question to "Not marked: a reflection, not a question", set "Layout" to "All on one page" and turn off "Shuffle the questions for each student".
  • Settings. "Minutes overall" at 10 leaves room for three takes of a 40-second extract and a listen to each; "Show the score" goes to "When released".
  • Recording. "Start recording" makes the browser ask for the microphone and nothing else. EdProcto asks for plain sound with no settings of its own, so the processing described above is the browser's and the phone's choice: hence your own test take. "Stop recording" sends the take, up to 25 MB, to EdProcto's server, and a player lets the student listen back; "Record again" replaces it. A take still running when the clock runs out is stopped and kept by itself, so stopping early is for listening back.
  • Marking. Until you mark it, a recording shows "to grade" and counts as zero. On the results page, "Grade" beside a name opens the submission: the player, your rubric, "Points" in half points, a "Note for the student" box and the "Save my grade" button. The listener gives the mark. After "Release the scores", each student opens the quiz's link on the same device and sees the score with your note.

On Free you get ten quizzes a month for thirty students each, on the core templates. All 111 templates, the recorded shapes included, come with Tutor at 100 students a quiz, and Educator raises that to 500 with the full analysis. See pricing. A recording is evidence of one take on one day, not proof of who played, and the score is yours to turn into a grade.

Quick questions

How long should a recorded performance be?

Under a minute for a class task: twenty-eight takes of 45 seconds are 21 minutes of listening, and of two minutes, 56.

Should students get one take or the best of three?

Up to three, inside a time limit: in Bergee (2007) the take hardly moved the score, and students in Beh et al. (2025) spent days chasing a perfect one.

Is a phone good enough for marking music recordings?

On the evidence for voices, yes for pitch, rhythm and phrasing; tone and dynamics need a test take of the real instrument, loud and soft.

Does EdProcto record video of the performance?

Sound only. In EdProcto's Sing or play shape the browser asks for the microphone, so the mark rests on what is heard.

References

Sing or play

A vocal or instrumental performance, recorded.

In the builder now

See Sing or play

Read next