Elysia Tools
Navigation
Workflow Playbook
Create sweep, stereo or measurement-excitation fixtures, check their format and expected signal behavior, and hand off parameters, review evidence and file hashes without claiming a measured acoustic response.
Hubs
This workflow serves audio-software developers and QA teams preparing reusable input files. Start with a case ID, expected behavior and acceptance method. Choose a mono sweep for a changing frequency, independent stereo signals for a channel case, or an excitation package for a later recording experiment. Do not pass a sweep through all three generators as though they were successive repair stages.
Write down sample rate, waveform, active duration, frequency endpoints, peak target, phase and fades. Supported periodic frequencies must remain below 0.45 times the sample rate. Lower playback-system gain before listening and avoid automatic or unattended playback. A digital level such as -18 dBFS does not specify sound pressure or a safe listening exposure.
Mono chirps support linear or logarithmic trajectories, including descending and equal endpoints. The measurement-package logarithmic branch instead requires increasing endpoints and at least one second. Its ZIP contains excitation.wav, parameters.json and README.txt, plus inverse.wav for the logarithmic branch. The inverse is a float32 analysis filter, not audio to audition.
For ordinary WAVs, retain the settings and returned metadata separately; do not assume they have the same ZIP layout. Active frame counts are rounded from duration times sample rate. Excitation lead and tail silence increase the total duration. MLS uses repetitions of a complete period of 2^order - 1 samples and ignores the ordinary duration and fade settings; the single-impulse branch has one nonzero active sample. Do not trim, fade or resample an MLS reference while continuing to call it the same fixture.
A fixed noise seed reproduces the generated case under the same implementation and settings, but left and right use different derived streams. Peak targets and conservative harmonic normalization are not fixed RMS targets. Save the measured metadata rather than promising an RMS value from the level control alone.
Probe excitation.wav or an ordinary WAV, not the ZIP as if it were an audio stream. Compare codec, sample rate, channel count and total duration with the effective settings. A waveform can reveal missing activity or a gross envelope mismatch, but a full-length thumbnail cannot prove one sample, precise fade timing or absence of every clip. Use an appropriate numeric check when those properties are acceptance requirements.
A spectrogram should show the intended sweep direction and coverage; its time/frequency resolution limits what it can resolve. For a stereo sanity case, generate equal-frequency, equal-level active sine channels and vary only the intended phase: aligned and inverted cases should give contrasting observations. Do not use silence, independent noise or different-frequency channels to infer a definitive polarity fault. Browser decoding, output routing and analysis settings must accompany any phase observation.
If a file disagrees with the plan, preserve it as rejected evidence and generate a new revision. Do not normalize or repair it silently, because processing changes the reference. Visual approval does not replace sample-accurate tests.
The handoff ledger should contain case ID, purpose, exact settings, generation metadata, file names, expected observations, actual review results, owner and unresolved checks. Include local WAVs or the untouched excitation ZIP, inspection images and final SHA-256 hashes. Hashes identify bytes; they do not certify signal correctness or equipment calibration.
An excitation package is not a recovered room impulse response. A separate authorized recording and analysis workflow must account for latency, drift, clipping, automatic processing and response-tail length. The supplied inverse is a finite-band approximate analysis filter, not a guarantee of a calibrated transfer function. Do not claim those later stages were performed by generating a ZIP.
Use the related channel-routing workflow when existing programme channels need correction, the audio-analysis workflow for broader source inspection, and the conversion workflow when changing delivery format. Keep any converted derivative separate: its hash and signal properties may differ from the approved reference.
Workflow playbook
Write the expected observation before generating anything. Choose mono sweep, independent stereo or measurement excitation, with explicit level, sample-rate and duration rules.
Create only the cases needed. Save returned metadata; retain the excitation ZIP and its parameters and instructions. Keep inverse.wav as analysis data, never a playback track.
Probe each WAV and inspect a waveform. Reconcile active frames, padding and MLS periods; unexpected channel counts or silent output must be explained before continuing.
Inspect a spectrogram for the intended frequency path. For an active matched stereo sine case, review expected relative phase; do not interpret a general correlation reading as a full channel or hardware test.
Use a mono chirp for a frequency trajectory, independent stereo settings for channel or relative-phase cases, and an excitation package only when a separate recording and analysis stage is planned. These are alternatives, not a mandatory processing chain.
Keep periodic frequencies below 0.45 times the sample rate. Distinguish active frames from lead/tail silence; MLS uses complete periods and ignores the ordinary duration/fade controls.
A waveform or spectrogram is visual evidence, not sample-accurate certification. Evaluate correlation only for suitable active stereo material; silence and independent noise do not establish polarity correctness.
Hash the final WAVs, excitation archive and saved evidence, then deliver a case ledger with parameters, results, owners and open questions. Preserve unmodified originals locally.