Belmont Labs · BL-013
LABCAST ANALYZER
Full-Spectrum Audio Analysis · Experiment 013
spectrum analyzer · 20Hz–20kHz · global capture
Click Start Capture, then choose a tab/screen and enable “Share audio”.
MODE — RATE — FFT 8192 RES 5.9 Hz/bin METER —
SPECTRUM · WAITING FOR SIGNAL dBFS · sine-calibrated · peak per column · max of L/R
0
-6
-12
-18
-24
-36
-48
-60
-72
-90
20501002005001k2k5k10k20k
MOMENTARY
—
LUFS · 400 ms · K-weighted
SHORT-TERM
—
LUFS · 3 s · K-weighted
TRUE PEAK
—
dBTP · 4× os · hold —
SAMPLE PEAK
—
dBFS · hold —
RMS
—
dBFS · 300 ms · crest —
CLIP EVENTS
0
0 /60s · ≥3 smp @ 0 dBFS
A/B LOUDNESS COMPARISON · ITU-R BS.1770-4 / EBU R128 IDLE
keys: A · B · H
Δ B−A
AVERAGE SPECTRUM · 1/3-OCTAVE━ A━ B hover for band values
LEVEL HISTORY · LAST 60s · fill = true peak · line = short-term LUFS waiting…
BAND ENERGY · — normalized · autosens · 48 bands
Capture chain: Global Capture uses the browser's tab/system audio share (getDisplayMedia). This path is not guaranteed bit-transparent — the OS/browser may resample or apply internal gain staging before LabCast ever sees the signal. Treat readings as a reliable relative/trending measure, not a calibrated lab reference, unless verified against known-good source material.
Metering path: every sample is metered exactly once on the audio thread (AudioWorklet), in 100 ms steps. Readings are independent of FFT size, the display channel mode, and background-tab timer throttling. L and R are always measured separately. If the page is opened from a local file, Chrome blocks worklet loading and the same metering code runs in a ScriptProcessor instead (METER shows COMPAT) — identical maths, but it runs on the main thread, so a heavily busy tab could drop audio blocks. Serve over HTTPS for the WORKLET path.
Loudness: ITU-R BS.1770-4 K-weighting, stereo (L=R=1.0). Momentary 400 ms, short-term 3 s, integrated with −70 LUFS absolute and −10 LU relative gates over 75%-overlapped blocks; Loudness Range per EBU Tech 3342. Verified against EBU Tech 3341/3342 test signals. For A/B comparison, only measurement windows that start after a segment begins are counted, so the previous station's audio can't leak in. 1 LU = 1 dB of loudness difference.
Peaks: True Peak uses 4× polyphase oversampling (BS.1770-4 Annex 2 method); accurate to ~0.01 dB on sustained content, may under-read isolated high-frequency transients by up to ~0.7 dB. Sample Peak is the raw sample maximum. Clip events = runs of ≥3 consecutive samples at/above the clip threshold, counted once per run.
RMS: unweighted, 300 ms, mean of channel powers. Convention: a full-scale sine reads −3.0 dBFS (AES17 would call it 0). Crest = 300 ms peak − RMS.
Spectrum: getFloatFrequencyData (Blackman window) offset by +13.56 dB so a full-scale sine peaks at 0 dBFS — this undoes Web Audio's 1/N scaling and the window's 0.42 coherent gain. Each column shows the highest bin it spans across L and R. Broadband content reads lower per bin at larger FFT sizes (−3 dB per doubling), so use the Average Spectrum for level comparisons. Display only — no readouts come from it.
Average spectrum: 1/3-octave bands (base-10 centres, 20 Hz–20 kHz). Band power = 2·Σ|X|²/mean(w²) with fractional-bin edge weighting: a sine reads its true level and noise is integrated across the band, so results don't depend on FFT size (verified: sine exact, pink noise flat within ~1 dB). Power-averaged over each segment with L/R power-averaged; frames are gated at −70 LUFS momentary and skipped until the FFT window has cleared the switch-over. Sampled from display frames (~60/s, ~1/s in a background tab) — unbiased either way. Low bands blur together at small FFT sizes (noted under the plot). Loudness-matched mode shifts B by the integrated-loudness difference so only tonal balance differs.
SETTINGS · BELMONT LABS · BL-013
Analyzer
Teal
Amber
Rose
Blue
Purple
Mono
Phosphor
Windows XP
Windows 3.11
Windows 11
Synthwave
Abyss
Starfield
Old Glory
Composition
CGA
EGA
HyperCard
Size
Mode
Style
Affects only the Band Energy strip below the spectrum (a CAVA-style visual aid). Peak / RMS / Clip readings are unsmoothed and unaffected by this setting.
Bins
Larger FFT = finer frequency resolution (narrower Hz-per-bin) at the cost of time resolution — transients smear across more analysis frames. 8192 @ 48kHz ≈ 5.9 Hz/bin, a reasonable default for general-purpose measurement. Use 16384 for detailed low-frequency work; use 4096 if you need faster transient response.
Reference
0 dBFS is true digital full-scale on the time-domain Sample Peak reading. This is a sample-value peak, not an oversampled/inter-sample true-peak measurement. The −1 dBFS option flags material that's getting close, which is useful if you want early warning before an intersample/lossy-encode overshoot could clip downstream.
BL-013 — Belmont Labs.

LabCast was originally a multi-source audio player; that role now lives at Intercom19. LabCast has been reoriented around its one genuinely distinct capability: a full-spectrum, dual-channel audio analyzer built for real measurement work, not just visualization.

See the Measurement Notes panel on the main view for an honest account of the capture chain's accuracy characteristics and limits.