752 Singers' First Vocal Assessments: What's Actually Weakest

Most advice about what beginner singers struggle with is based on what experienced teachers remember noticing. That is valuable, but it is memory, and memory is selective. We had a different kind of evidence available: every singer who completes a first vocal assessment on Bloom Vocal gets scored on five areas by the same automated analysis, using the same criteria, in the same order.

So we counted.

What we measured

Between 2026-03-29 and 2026-08-10, 752 singers completed at least one AI vocal assessment, producing 1,063 assessment sessions in total. Each assessment scores five areas out of 20:

• Breath support — whether the sound is powered by steady airflow rather than throat pressure
• Tone and register — timbre consistency and control across register transitions
• Pitch accuracy — how closely sung pitches match intended ones
• Rhythm — timing relative to the underlying pulse
• Expression — dynamic and phrasing variation

For the distribution below we used one assessment per singer — the first one. This matters: if we pooled all 1,063 sessions, the singers who practice most would count several times and the picture would drift toward regular users rather than describing singers in general.

The result: breath support scores lowest on average

n = 752 singers, first assessment only.

The spread between the weakest and strongest area is 1.78 points — real, but not dramatic. What is more interesting is the standard deviation column. Rhythm has the narrowest spread (3.09) while pitch accuracy has the widest (3.81). Singers resemble each other on rhythm far more than they resemble each other on pitch.

But individual weak points tell a different story

A low group average does not mean "this is the area most people are worst at." Those are two different questions, and in this dataset they give two different answers.

We took each singer's five scores and identified their single lowest area. One complication: 34.4% of singers had a tie — two or more areas sharing the same lowest score. Assigning ties arbitrarily would inflate whichever area happens to sort first, so we excluded them and counted only the 493 singers with one unambiguous low point.

n = 493 singers with a single lowest-scoring area (ties excluded).

So: breath support drags the average down across the whole group, but tone and register is the area most often at the bottom of an individual singer's own profile.

Both statements are true, and the difference is not a technicality. Breath support has a low average partly because it is broadly mediocre — lots of singers sit somewhat below the middle. Tone and register produces more singers who are specifically and distinctly worst at it. If you are building a practice plan for a room full of singers, breath support is the safer group bet. If you are diagnosing one singer, register transitions are the more likely single culprit.

Rhythm is the consistent floor

Rhythm was the strongest area on both measures: highest average (13.72) and rarest weak point (7.5%). Combined with its narrow spread, that suggests rhythm is where untrained singers are most uniformly competent.

The practical reading: if a singer's rhythm score comes back low, it stands out against a strong baseline and deserves attention. A mediocre breath support score is ordinary; a mediocre rhythm score is not.

What this data does not show

We want to be specific about the limits, because a number is only useful if you know what it cannot support.

It is not a study of improvement. This is a snapshot of first assessments. We separately examined before-and-after change and found the sample too thin to publish: only 41 singers had two or more assessments, with an average of 8.6 days between the first and most recent. Vocal technique does not meaningfully change in 8.6 days, and 41 people is not enough to separate signal from noise. We would rather say nothing than publish a number that cannot bear weight.

The group is self-selected. These are people who chose to download a vocal coaching app and complete an assessment. That is a motivated group, and motivated groups differ from the general population in ways that are hard to correct for.

It describes beginners. 614 of 752 singers self-reported as beginners. The intermediate group was 7 people — far too few to report separately, so we have not. Read this as a portrait of beginner voices.

Scores come from audio analysis, not a teacher in the room. Automated analysis captures what is measurable in a recording. A human teacher can see jaw tension, watch breathing, and hear things that a scoring model does not represent. These scores are a structured starting point, not a replacement for that.

It is observational. Nothing here establishes cause. We are describing what we counted.

Why we are publishing the method alongside the numbers

Two decisions in this analysis changed the headline, and both would have been invisible if we had only published the result.

The first is the tie handling. Had we assigned each tied singer to whichever area sorted first alphabetically, breath support would have appeared as the weakest area for 49.3% of singers — a much more quotable number, and a wrong one. It would have been an artifact of sorting, not a finding about voices.

The second is using one assessment per singer instead of all 1,063. Pooling sessions would have overweighted the most active users.

Neither choice is exotic. Both are the difference between a number that describes singers and a number that describes our data pipeline.

If you want to check your own five areas

The same five-area assessment is what Bloom Vocal runs on a recording of your voice. If you want to see how the areas relate to each other before testing anything, our guides on breath support and appoggio and resonance and register placement cover the two areas that this data flags most often.

For a broader view of what different tools measure and what they cost, see the best vocal apps in 2026 and how much a vocal coach costs.

Method summary

• Source: Bloom Vocal production database, aggregate queries only. No personally identifying information was accessed or output.
• Window: 2026-03-29 to 2026-08-10.
• Population: 752 singers with at least one completed AI vocal assessment; 1,063 sessions total. Internal test and developer accounts excluded.
• Distribution basis: first assessment per singer (n = 752).
• Weak-point basis: first assessment per singer, ties excluded (n = 493 of 752; 34.4% had tied lowest scores).
• Scale: each area scored out of 20; observed range 1–19. Total score out of 100; observed range 12–97.
• Self-reported experience: 614 beginner, 7 intermediate, 131 unspecified.