Input Recording

Input Recording isn't tied to a specific calibration procedure. It's a general-purpose tool for capturing and reviewing whatever's coming into one or more input channels simultaneously.

Comparing with Measure Sound

If you're comparing results with the lab's MATLAB program, Measure Sound, read Comparing results with Measure Sound. The two programs calculate some levels differently, dBA in particular, and that page explains how to allow for it.

Recording

Run CFTSCal and select the Input Recording workspace.

Two recordings plotted, one from each ear's microphone. The shaded band on the waveform is the selected region; the Analysis table and the spectrum below it are computed from that part of each recording only (see Selecting a region).

Field	What it means
Number of inputs	How many channels to record at once, from 1 up to the total number your hardware exposes. One row appears below for each.
Input (first dropdown on each row)	Which physical input this row records. The dropdown doubles as the row's label, so there's no separate "Input 0"/"Input 1" text. Pick any channel here; the same physical channel can't be assigned to two rows at once.
Sensor type and calibration (next one or two dropdowns)	What kind of sensor is attached (see Sensor types below), plus — for most types — a second dropdown picking which specific calibration to use.
dB gain (last dropdown on each row)	The preamp gain, in dB, currently applied to that row's channel. It must match the amplifier's actual setting.
Target folder	Which folder the recording is saved in. Shared across all rows, since they're saved together as a single recording. Leave it at (auto create from generator) to have a folder made for you, named after the Generator text (lowercased, with spaces and punctuation turned into hyphens).
Generator	Free text describing what produced the sound — e.g. pistonphone, speaker 3. It's saved with the recording exactly as you type it, but isn't remembered between sessions.

Sensor types

Type	What it means
Meas. Mic., Generic Mic., Input Amp., Starship	Loads a real, on-file calibration — pick the specific one from the second dropdown. This calibration converts the recorded voltage to Pascals.
Unity	A pass-through: records the raw voltage without converting it to Pascals. Since there's nothing to pick, the second dropdown is hidden.
Nominal	For a device with no measured calibration on file — e.g. going off a spec sheet value. Replaces the second dropdown with an mV/Pa field where you type the sensitivity directly; the recording is calibrated to Pascals using that value. Note the direction of the ratio: millivolts generated per Pascal, the convention used on most datasheets. If your datasheet gives Pa/V instead, take the reciprocal — see Which way round is "sensitivity"?

Running the recording

To run the recording, click the Record button — this captures every active channel simultaneously, saved together as one recording. The button stays disabled until:

• every row has a sensor selected,
• no two rows point at the same physical input, and
• the Generator contains at least one letter or digit.

If that state is somehow reached anyway, clicking Record shows a warning explaining why rather than failing silently.

Reviewing the results

Select one or more recordings in the Recordings list to plot them. The Input Recording plot shows the calibrated time-domain waveform; the region you select drives both the PSD plot and the Analysis table below. If a recording is longer than 90 seconds, the plot opens on its first 90 seconds; drag the plot to move along it, or zoom out (scroll, or right-click > View All) to see all of it.

Recordings (the list) shows every recording ever made for this workspace, with the following columns:

Column	Meaning
Name	Which folder the recording is filed under (with auto create, named after the generator).
Date	When the recording was made.
Generator	Which stimulus/generator label was used.
Channel	Which input channel(s) were recorded — comma-separated if more than one.
Sensor	Which sensor was attached to each channel — comma-separated in the same order as Channel.
Gain	The preamp gain, in dB, that was in effect for each channel — comma-separated in the same order as Channel.

A warning icon after a recording's name means it couldn't be read properly (hover for why); a sticky-note icon means it has a note. To add or change a note, right-click the recording and choose Edit note… — the note is saved inside the recording's folder, and hovering over the row shows it.

Each recording gets its own color, matching its highlight in the Recordings list. If a recording has more than one channel, all of its channels share that color but are distinguished from each other by line style (solid, dash, dot). The Analysis table has a column for each recorded channel, headed by the channel and the date and time of its recording, so channels you're comparing (say, the left and right ears) sit side by side. The strip of color at the top of each column matches that recording's color, and a recording's channels appear as neighboring columns. Each row is one measurement of whatever's inside the currently selected region:

Row	What it shows
Dur (s)	Length of the selected region.
dB SPL	RMS level, after the Filter.
dBA	A-weighted RMS level. Always measured from the unfiltered recording, whatever the Filter is set to, so it's there without switching the filter to dBA. It's A-weighted exactly as the dBA Filter would be, following the Zero-phase and Exact A-weighting checkboxes.
peSPL	Peak-equivalent SPL: the RMS level of a sine wave with the same peak-to-peak amplitude. For a pure tone it equals dB SPL; for clicks and other brief sounds it's the usual way to state their level.
Peak SPL	Half the peak-to-peak amplitude, in dB SPL — 3 dB above peSPL. (This was called "pe SPL" before peSPL was added.)
Max (Pa), Min (Pa)	The most positive and most negative pressure. A lopsided pair shows the waveform isn't symmetric, which the peak-to-peak rows can't.

Every row except dBA is measured after the Filter. A dash means the selected region doesn't overlap that recording.

Selecting a region

• Ctrl+drag anywhere on the time plot to draw a new region from scratch — including inside the current region, so you can select a smaller part of it.
• Drag an edge of the existing region to resize it.
• Drag the middle of the region to move it without resizing.
• A plain drag (no Ctrl) pans the plot as usual.

Only the samples inside the selected region are used for the Analysis table and PSD.

Choosing a filter

The Filter dropdown controls the filtering that gets applied to the signal before it's plotted and the level is computed. It only changes what's shown and measured; the saved recording is never altered.

Mode	What it does
High-pass (default)	Removes everything below a Cutoff frequency (20 Hz unless you change it). With the default order of 3 it matches the HP filter on a GRAS 12AQ power module: a 3-pole Butterworth, 3 dB down at the cutoff and falling 18 dB per octave below it. Use it to keep building rumble and handling noise out of the level. A higher order cuts off more steeply. Like the 12AQ's own filter, it shifts the phase of low frequencies slightly, unless Zero-phase is ticked.
Unfiltered	No filtering at all — deliberately not labeled "dBZ", since that would imply a standardized flat response over a defined range, and this is simply whatever bandwidth the raw recording happens to have.
dBA	Standard A-weighting (IEC 61672-1). See Exact A-weighting below.
Band-pass	Keeps only a band of frequencies: centered on Center freq., width octaves wide (1/3 octave unless you change it), with an adjustable order (higher orders roll off more sharply outside the band). The band edges are 3 dB down. A center of 1000 Hz and a width of 4 octaves (250 Hz to 4 kHz) has the same band edges as the band-pass filter in the lab's older MATLAB program, Measure Sound, but not the same shape.

Two checkboxes at the end of the Filter row change how the filter is applied. Neither changes the filter's shape (its cutoffs, slopes or band edges), so neither changes the dB SPL or dBA of a steady sound:

• Zero-phase (off by default). Off, each filter acts the way an analog filter would, like the 12AQ's high-pass or a sound level meter's A-weighting: frequencies near the cutoff come out slightly delayed, which reshapes clicks and other brief sounds, so their peSPL, Peak SPL and Max/Min can differ from the unfiltered recording. Ticked, the same filter is applied without any delay, so the filtered waveform lines up exactly with the original. The catch is that a zero-phase filter responds a little before a sudden sound starts, which can look like an artifact just ahead of a click. Leave it off to match what analog equipment would read. Tick it to compare waveform shapes or timing.
• Exact A-weighting (off by default). Off, A-weighting uses a standard digital approximation of the curve. It's exact at low and middle frequencies but falls short of the standard near the top of the recording's frequency range: at a 48 kHz sampling rate it's 1.2 dB low at 10 kHz and 6.4 dB low at 16 kHz (at 96 kHz, only 0.3 and 1.1 dB). Ticked, the exact curve is applied to the recording's spectrum, which is correct at every frequency but takes longer (a second or two for a long recording), and with Zero-phase off it responds very slightly before a sudden sound starts. Tick it when sounds above about 8 kHz matter and you're sampling at 48 kHz or below.

A third checkbox, Emulate Measure Sound (off by default), is only for comparing with results from the lab's older MATLAB program, Measure Sound. Ticked, Input Recording calculates everything as Measure Sound does, including its correction factors on the first two channels, and the plot, Analysis table, Play region and Export region all use that emulated signal. A note above the Analysis table shows while it's ticked. Leave it unticked for new measurements.

How Zero-phase works

Every filter has a frequency response, \(H(f)\): how much it lets through at each frequency \(f\) (its magnitude, \(|H(f)|\)), and how much it delays each frequency (its phase). The delay is what reshapes brief sounds, so a zero-phase filter has to keep the magnitude and drop the delay.

The usual way to do that is to run the filter once forwards and then once backwards over the result (filtfilt in MATLAB and SciPy, or SciPy's sosfiltfilt). The backwards pass undoes the delay, but it also applies the magnitude a second time. Every attenuation in dB doubles: a cutoff designed to be 3 dB down comes out 6 dB down, the slope is twice as steep, and A-weighting would turn 50 Hz down by 60 dB instead of 30 dB. The filter you get is no longer the one you chose, so the levels change too.

Input Recording does it differently. It takes the recording's spectrum (its FFT), multiplies each frequency by \(|H(f)|\) of the filter you chose (exactly once), and converts the result back into a waveform. The magnitude is applied once, so cutoffs, slopes and band edges are exactly as designed, and dB SPL and dBA are the same as with Zero-phase off. Only the delay is gone.

This has two small costs. It takes longer, about a second or two per channel for a long recording. And a spectrum treats the recording as if its end were joined to its start, so Input Recording pads both ends with extra samples, long enough for the filter to settle, and trims them off afterwards; otherwise sound from the end of a recording would leak into its beginning.

Listening to and exporting a region

The Play region and Export region… buttons, at the right end of the Filter row, work on the selected region exactly as it's plotted and measured — after the Filter — so what you hear and save is what the Analysis table describes.

• Play region plays one channel through the computer's default sound output. Click a cell in that channel's column of the Analysis table to choose it (otherwise the first column plays). Click the button again (it now reads Stop) to stop early. Playback uses a fixed scale, 20 Pa (120 dB SPL peak) at full volume, so recordings can be compared by ear: a 94 dB SPL calibrator tone plays fairly quietly, and very quiet recordings may be hard to hear. If the region peaks above 20 Pa you're warned that it will be distorted, and can choose whether to play it anyway.
• Export region… asks for a folder and saves the region of every ticked recording there as a WAV file, one per recording, named after the recording and the region (e.g. 20260616-101544_18.000-22.000s.wav). The files use the same calibrated format as Export as WAV… — see Exporting a calibrated WAV — and also record the region and the Filter settings. An existing file is never overwritten; _2, _3, … is added to the name instead.

Both buttons are greyed out until the region overlaps a ticked recording.

Reviewing PULSE .mat files

Input Recording can also show recordings made with B&K PULSE and saved as MATLAB .mat files: the files the lab's MATLAB program, Measure Sound, opens, including the ones its Save button writes. There's no Open button. Instead, copy the files into Input Recording's folder, input-recording inside the calibration folder (Documents/cftscal/input-recording unless you've changed it in Workspace > Settings), and they appear in the Recordings list beside the recordings made in cftscal:

• A file inside a folder is listed under that folder's name, so you can file .mat recordings in the same folder as cftscal recordings of the same thing (for example input-recording/BK-4123/).
• A file placed directly in the input-recording folder is listed under its own name.

They're plotted, measured, played and exported just like cftscal recordings. Things to know:

• Levels. The samples are taken to be in pascals, exactly as stored in the file, with no correction factor applied.
• Columns. Channel shows each channel's name from the file (e.g. LeftEar), or Channel 1, Channel 2, and so on if it has none, and Sensor reads .mat file (Pa). Generator shows the file's name, since PULSE files don't record what made the sound, and Gain is blank.
• Date. PULSE files store the date they were recorded, but not the time. The time comes from when the file was saved, if that was on the same day. Otherwise the date is shown at midnight. A file with no date at all shows when it was saved, or failing that, when it was last changed (which copying it may have altered).
• Notes can't be added to a .mat recording, since there's nowhere in the file to keep them.
• Delete deletes the .mat file itself.
• A file that can't be read, or isn't a PULSE recording, is still listed but marked with a warning icon; hover over it to see why. Files saved in MATLAB's newer v7.3 format can't be read. Open them in MATLAB and save them again with save(filename, '-v7').

Interpreting the levels

A single number describing a broadband signal's level always depends on how wide a band you're talking about, so it's worth knowing which quantity you're looking at:

• Spectrum level is the level in a 1 Hz-wide band — i.e. the level per hertz. This is what the PSD plot shows, at each frequency.
• Band level is the total level integrated over a band of width \(\Delta f\). This is what the Analysis table's RMS dB SPL reports, over whatever bandwidth the filter leaves in place.

They're related by:

\[ BL = SL + 10 \times log_{10}(\Delta f) \]

Worked example. A flat noise with a spectrum level of 56 dB spanning 4–64 kHz has a band level of \(56 + 10 \times log_{10}(60000) = 103.8\) dB SPL. Note the \(10 \times log_{10}\) — this is a power ratio, unlike the \(20 \times log_{10}\) used for amplitude ratios elsewhere.

An unfiltered level often measures your noise floor

Because band level grows with bandwidth, a low noise floor spread across a wide bandwidth can dominate the reported total. For a 4 kHz-wide signal band at a 65 dB spectrum level (a 101 dB SPL band level) sitting on a flat noise floor that extends out to 50 kHz:

Noise floor (spectrum level)	Reported total
30 dB	101.0 dB SPL
40 dB	101.2 dB SPL
50 dB	102.4 dB SPL
60 dB	107.7 dB SPL

A noise floor 5 dB below the signal's spectrum level still adds 6.7 dB to the total, purely because it's 11 dB wider in bandwidth. This is why the region selection and the filter matter: restricting the analysis to the region and band you actually care about is what makes the reported level a measurement of your stimulus rather than of your noise floor. Total level is easy to get wrong has the full worked numbers.

Exporting a calibrated WAV

Right-click any recording (in Input Recording, or any other workspace's list) and choose Export as WAV… to save it as a standard WAV file, calibrated so that a sample value of 1.0 represents exactly 1.0 Pascal of sound pressure (with the exception of recordings generated by unity, in which case the output is 1.0 represents exactly 1.0 Volts). If the recording has multiple channels, they're exported as a single interleaved multi-channel WAV file, all channels sharing the same sample rate. The file also carries the recording's metadata (per-channel sensitivity, sensor ID, calibration date, etc.) embedded directly in it, in a way that doesn't interfere with normal playback.