MastertakeProduction coach and session notebook

How it works

From a bounce to a plan you can act on

Mastertake follows the way a patient engineer coaches someone new: ask what you’re going for, measure, explain, suggest one thing at a time, and write down what you decided.

The flow

  1. Intent card

    Three to five quick questions at the start of a session or project: what should this song feel like, who is it for, what worries you, what gear and room do you have, and what’s your level. About two minutes, so advice is never given blind.

  2. Mix Check

    You drop in a bounce. It is analyzed on your own computer and turned into a small file of numbers. The AI model reads those numbers together with your intent card and the playbook, explains what they show, and suggests at most three things to try, in order.

  3. Debrief and session note

    At the end of a session, a few questions: what did you decide, what’s still open, what will you try next. The session note is written from your answers and any short marks you typed along the way, never from recorded conversation.

  4. Next-step plan

    A short plan: what to try, in what order, and how to check it worked. Next time, Mastertake remembers what you already tried.

What reaches the AI model

Sent, only with your consent
The measurements, your intent card and the notes you choose to share.
Never sent
Your audio. And nothing is recorded in the room.

A sample Mix Check report

Sample only: not a real artist’s track, and not advice

The measurements below come from a synthetic test signal, not from music anyone made. The intent card is fictional, written to show the format.

Sections 3 and 4 only illustrate how suggestions will be laid out. The advice rules are still being written and reviewed by our studio, so nothing in this sample is advice.

Mix Check · sample report

synthetic-demo-bounce.wav

Format
WAV · PCM_24
Sample rate
48 kHz
Channels
2 (stereo)
Length
30.0 s
Analyzer
mastertake-analyze 0.1.0
Analyzed
Fingerprint (SHA-256)
2d6c41b51325bab8…

1. What you told us

Fictional intent card, written for this example.

What should it feel like?
Warm and close, as if the singer is in the room with you. Not loud just for the sake of it.
Who is it for?
A first single, for friends and a small local audience.
What worries you?
The bass sounds big on my headphones but disappears on my phone.
Gear and room
A laptop, a small two-input audio interface, closed-back headphones, and a pair of small monitors on a desk in a bedroom.
Level
First year of mixing my own songs.

2. What we measured

Measurements of the synthetic test signal, with a plain-language explanation under each value.
MeasurementValue
Loudness
Integrated loudnessloudness.integrated_lufs−15.3 LUFS
The average loudness of the whole file, in LUFS, a standard unit for how loud audio sounds to people. Numbers closer to zero are louder.
Loudness rangeloudness.loudness_range_lu1.8 LU
How much the loudness moves between the quieter and louder parts of the song. A small number means the level stays fairly even.
Loudest 3-second stretchloudness.max_short_term_lufs−14.3 LUFS
The loudest moment when loudness is averaged over 3-second windows (short-term loudness).
Loudest instantloudness.max_momentary_lufs−13.2 LUFS
The same, over much shorter 0.4-second windows (momentary loudness). It catches brief loud hits.
Peaks
Sample peakpeaks.sample_peak_dbfs−0.4 dBFS
The highest single sample in the file. 0 dBFS is the digital ceiling; nothing can go above it.
True peakpeaks.true_peak_dbtp−0.4 dBTP
An estimate of the highest level between the samples, which is what a converter or an encoder can actually produce. It can be higher than the sample peak.
Clipped runspeaks.clipped_sample_runs0
How many places several samples in a row sit at the digital ceiling: a typical sign of clipping, where the waveform is cut flat.
Dynamics
Peak-to-loudness ratio (PLR)dynamics.peak_to_loudness_ratio_db14.9 dB
The gap between the true peak and the integrated loudness. A smaller gap usually means the mix has been compressed or limited harder.
Crest factordynamics.crest_factor_db13.9 dB
The gap between the peak level and the average (RMS) level: another view of how far the loudest hits rise above the body of the sound.
Tonal balance
Each band’s share of the mix’s total energy, in dB. Closer to zero means a bigger share.
Sub · 20–60 Hzspectrum.bands.sub.share_db−1.8 dB
The deepest bass, felt as much as heard. Phone and laptop speakers play very little of it.
Bass · 60–250 Hzspectrum.bands.bass.share_db−6.1 dB
The body of the bass and the kick drum, and the low notes of most instruments.
Low mids · 250–500 Hzspectrum.bands.low_mid.share_db−12.2 dB
Warmth and fullness. Many instruments overlap here, so it can get crowded.
Mids · 500 Hz–2 kHzspectrum.bands.mid.share_db−15.3 dB
The core of most instruments and voices. Small speakers reproduce this range well.
High mids · 2–6 kHzspectrum.bands.high_mid.share_db−42.5 dB
Presence and clarity: consonants, pick attack, the crack of a snare. Ears are especially sensitive here.
Highs · 6–20 kHzspectrum.bands.high.share_db−28.5 dB
Air and sparkle: cymbals, breath, the sound of the room.
Stereo image
Stereo correlationstereo.correlation0.20
How alike the left and right channels are. +1 means identical (mono) and 0 means unrelated; negative values mean the channels work against each other and can cancel out when played in mono.
Side-to-mid levelstereo.side_to_mid_db−1.8 dB
How loud the stereo difference (side) is compared with the centre (mid). Closer to zero means a wider image.
Low-end correlation (below 120 Hz)stereo.low_end.correlation0.13
The same correlation, measured on the low end only. Values near +1 mean the bass is close to mono.
Low-end side-to-mid level (below 120 Hz)stereo.low_end.side_to_mid_db−1.2 dB
How much stereo width there is in the low end. More negative means a narrower, more centred low end.
Noise
Noise floornoise.noise_floor_dbfs−78.0 dBFS
The level of background sound (hiss, hum, room noise) in the quietest parts. More negative means quieter.
Silence
Silence at the startsilence.leading_s0.50 s
How long the file stays below the silence threshold (−60.0 dBFS) before the sound starts.
Silence at the endsilence.trailing_s2.41 s
How long the file stays below the same threshold after the sound ends.

A dash (—) means the value wasn’t measured, for example stereo values on a mono file.

3. What it suggests

Illustration of the format only. The rule IDs are drafts and their thresholds haven’t been set, so these are not recommendations.

  1. Priority 1
    Check whether the bass survives in mono
    Rule
    PB-002 (draft)
    Cost
    Free
    Measurements cited
    • stereo.low_end.correlation = 0.13
    • stereo.correlation = 0.20
    Why
    You said the bass disappears on your phone. Most phones play a single channel. The low-end correlation is far from +1, which means the left and right bass are different enough to partly cancel when they are added together.
    Try
    Switch your DAW’s master to mono and play the chorus. If the bass drops away, bypass any stereo widener, chorus or doubling on the bass and kick, then listen again.
    How to check
    Run a new Mix Check: the low-end correlation should move toward +1. Then play the chorus on your phone again.
  2. Priority 2
    Leave a little room at the top
    Rule
    PB-001 (draft)
    Cost
    Free
    Measurements cited
    • peaks.true_peak_dbtp = −0.4 dBTP
    • peaks.clipped_sample_runs = 0
    Why
    The true peak sits just under the digital ceiling. Peaks that close can distort after conversion or streaming encoding, and your intent card says you don’t want it loud for its own sake.
    Try
    Pull the master fader or the limiter’s ceiling down a little, instead of adding another plugin.
    How to check
    Bounce again and compare the true peak in the next Mix Check. Then A/B the two bounces at matched loudness.

4. How to check it worked

  • Change one thing at a time, so you know what made the difference.
  • Bounce again and run a new Mix Check. Compare the two reports side by side.
  • Listen at matched loudness (A/B), so the louder version doesn’t win just for being louder.
  • Try a second speaker: a phone, a laptop, a car or headphones.
  • Compare with a reference track you own and like.

5. What we cannot know

I can’t hear this track. Here is what the measurements show.

Numbers can’t tell whether a part is musically right, whether the vocal moves anyone, or whether the mix matches your taste. They don’t know your room or your speakers either. That’s why every report starts from your intent card, and why taste is always labelled as taste.

6. If you want a human

When the measurements and the playbook run out, Mastertake helps you prepare for a session at our studio: a short summary of your project, and you choose what to share.

Want to try it on your own tracks?

The first group is small, for adults, and works with us at our studio in Izmir.

How to join