Manual Measurement Explained

Omnissiah has two ways in. Room correction is the guided one: it measures, analyses, calculates filters and offers to apply them. Manual Measurement is the other, reached from Calibrate in the sidebar and then Manual Measurement on the setup page, and it does none of that. It plays pink noise and shows you the result, continuously, and leaves every conclusion to you.

Transfer function

The measurement behind the display. What the microphone heard is compared against a reference copy of what was sent, so the test signal's own spectrum cancels and what remains is the behaviour of the speakers and the room in between.

That makes it the screen for questions rather than fixes. Whether the left and right speakers match, whether that dip is the desk or the speaker, whether moving the sub 200mm helped, and whether a correction you already applied did what it claimed. When you have the measurements you need, Create Calibration hands them to the guided flow, so the manual work feeds the automatic work rather than competing with it.

Start Noise plays pink noise through the selected output and the charts begin to fill. Stop Noise ends it. Auto Capture does the whole run without supervision: it starts the noise, waits until the measurement is stable, takes a snapshot and stops, showing its progress as a percentage while it waits.

Three things are drawn. Magnitude, roughly 25Hz to 20kHz on a log scale, is the shape of the sound, with peaks as too-loud regions and dips as too-quiet ones. Phase sits below it, wrapped between plus and minus 180 degrees, blanked rather than drawn as noise wherever the reading is not trustworthy. Running through the magnitude chart is the coherence trace, coloured red through yellow to green.

Coherence

A score from zero to one, computed per frequency, of how consistently the microphone signal followed the reference. Green means the measurement is reporting your speaker. Red means noise, reflections or timing problems dominated that band, and the magnitude and phase drawn there are not evidence.

Read coherence before anything else, because it tells you which parts of the curve you are allowed to use. Hovering the chart gives Freq, Gain, Phase and Confidence at that point, Confidence being the coherence as a percentage. Aim for mostly green across the range you care about; red everywhere means raising the playback level, quietening the room, or checking the microphone, and a low-confidence measurement misleads the correction engine exactly as it misleads you.

The gear icon in the toolbar opens the measurement settings. Two of them change the measurement and three change only the picture, and keeping that straight is most of the skill.

Averaging comes in two families. Slow, Fast and Ultra-Fast behave like a level meter, easing continuously toward the latest sound and settling over roughly 1.5 seconds, 500 milliseconds and 250 milliseconds. FIFO4 to FIFO32 average exactly the last 4 to 32 frames instead, so the trace settles in a predictable time and then holds still if the room does. The dropdown is locked while noise is playing, so choose before you start.

Smoothing defaults to a sixth of an octave, which is roughly what the ear resolves and where most judgements are made. Widening it to a third or a full octave shows tonal tilt and hides narrow features; narrowing it toward None shows every sharp dip, including the ones that move when you move your head. It always makes narrow dips look shallower than they are, so never size a boost from a heavily smoothed trace.

The Coherence setting hides frequencies whose confidence falls below the level you pick. Raising it shortens the trace, usually at the bottom, and that is the point: a shorter curve that is true beats a complete curve with invented regions. Clock Drift Correction is narrower still. It matters only when the microphone input and the speaker output run on different interfaces, because two clocks running at slightly different rates make the measured latency walk and the phase trace rotate on its own. On one interface, leave it disabled.

A snapshot freezes the live measurement so it can be kept, named, coloured, assigned to an output and compared against others. The Realtime row at the top of the list is always the live signal; each captured snapshot sits below it with a colour dot to show or hide it on the chart, a name to select it, and a cross to delete it. Left against right, before against after moving a speaker, sub against mains: the comparison is the measurement.

Selecting one opens a bar of controls over it, and the useful ones are the manual offsets. Delay steps in 0.5ms, gain in 0.5dB, and polarity is a single flip, all applied to the chart and not to the device. That means you can test an alignment before committing to it: nudge the subwoofer snapshot's delay until it stops fighting the mains through the crossover, read off the number, and only then set it on the output.

Set as Reference marks one snapshot as the thing everything else is measured against, which is what the phase tuner needs. Combine sums two or more snapshots into a single response, which is how you see what a sub and a main do together rather than separately. Export CSV writes the selected snapshot out with its name, colour, sample rate, latency and offsets, Export All does the lot, and Load brings them back, so a measurement session survives the app closing.

A wrapped phase trace is hard to read and harder to compare, so the phase plot has a Trace and Tuner switch. In Tuner mode each visible snapshot becomes a needle on a scale running from minus 180 to plus 180 degrees, showing how far that trace sits from the reference, with a marked zone in the middle for within 15 degrees and the needles coloured by how close they are. Below it, a second pane shows agreement against frequency, so you can see where the two traces agree rather than only how much they agree overall.

The reading is weighted by the inverse of frequency, so every octave carries roughly equal weight. That matters because 10 degrees at 40Hz is a hundred times more absolute time than 10 degrees at 4kHz, and because a single microphone position is least representative in exactly the top octaves that would otherwise dominate an unweighted average.

The workflow that falls out of it: capture the mains, capture the sub, set the mains as the reference, then step the sub snapshot's delay offset and watch the needles converge. Check polarity first, since a polarity error shows as a constant half-cycle offset across the whole overlap while an arrival-time error rotates progressively with frequency. When it is as tight as it will get, the delay you dialled into the snapshot is the delay to set on the output.

Below the charts sits a diagnostics panel running the same room analysis as the results page, but live. It writes plain-language cards about reflections, room nulls, resonances and measurement-quality problems as the microphone hears them, so moving a speaker or quietening the room updates the findings while you stand there.

Hovering a card shades its frequency band on the chart, which answers the question a card alone cannot: where exactly. Selecting a snapshot switches the panel to describing that capture instead of the live signal. This is the part that turns a curve into a decision, because it separates the dips a filter can fix from the ones that are a reflection and will not respond to any amount of boost.

Everything on screen is true at the microphone and progressively less true away from it. Low-frequency behaviour varies by many decibels within a seat width, so a response measured at one point is not the response of the room. Capture several positions around the listening area, give them names you will still understand tomorrow, and act on what they agree about.

A magnitude curve also says nothing about decay. A mode that rings on after the note has stopped still rings after it has been cut, and the cut is worth having without being the fix. Bass trapping changes decay, filters change level, and a flat measurement is not evidence that the first problem was addressed.

This article covers what the screen is for and how to read it. The user manual documents every control on it, with screenshots, and is the place to go for the specifics.