Every card states one bounded question, the observed result and what the test does not establish. *Where shown, timings describe one run on the test machine; they are not speed guarantees.
Question: Can the tool analyse an intentionally off-pitch monophonic fixture, correct it to C major and export a readable WAV?
Observed: 42 of 45 analysed segments were corrected to C major and the exported 333,272-byte WAV passed the file check.
- Audio payload requests
- 0
- External requests during workflow
- 0
Scope limit: This fixture tests one short synthetic monophonic voice. It is not an accuracy score for every singer, key or recording condition.
Question: Can the tool add a front-cover picture while preserving the compressed FLAC audio payload byte for byte?
Observed: The exported FLAC contained a type-3 front-cover picture and retained the original compressed audio payload byte for byte.
- Audio payload requests
- 0
- External requests during workflow
- 0
Scope limit: The fixture covers one valid FLAC structure and one front-cover replacement path, not every malformed or unusual metadata layout.
Question: Can the tool load a local pad and export a readable one-pad JSON board backup?
Observed: The six-second fixture loaded into one pad and the exported one-pad JSON backup passed the structure check.
- Audio payload requests
- 0
- External requests during workflow
- 0
Scope limit: This result covers one pad and one backup export. Storage capacity and playback latency still depend on the browser and device.
Question: Can the granular instrument load a local fixture, render a short capture and export a readable WAV?
Observed: The tool captured 0.7 seconds of granular output and exported a valid 138,284-byte WAV.
- Audio payload requests
- 0
- External requests during workflow
- 0
Scope limit: Granular scheduling is intentionally stochastic, so this verifies the workflow and file output rather than a fixed artistic result.
Question: Can the analyser load a local audio fixture, draw its spectrum and export a readable PNG?
Observed: The six-second fixture loaded and the visible spectrogram exported as a valid 10,037-byte PNG.
- Audio payload requests
- 0
- External requests during workflow
- 0
Scope limit: This published case covers the live analyser viewport. The separate full-file mode has a targeted regression test, but neither mode is a calibrated laboratory spectrometer or SPL meter.
Question: Can the real UI recover controlled 80, 100, 128 and 160 BPM pulse trains and disclose its intentional half/double-time folding at 60 and 200 BPM?
Observed: All 6 cases passed. Exact grid: 80→80 BPM, 100→100 BPM, 128→127.7 BPM, 160→160 BPM. Declared half/double-time interpretations: 60→120 BPM, 200→100 BPM.
- Audio payload requests
- 0
- External requests during workflow
- 0
Scope limit: These fixtures are steady synthetic click grids. Swing, syncopation, rubato, changing tempo and mastered music remain outside this wave.
Read this suite's machine-readable cases →
Question: Can the browser-resampled analyser return the exact authored tonic, major/minor mode and Camelot code for six deterministic chord progressions?
Observed: All 6 authored progressions matched exact tonic, mode and Camelot code: C major (8B), D major (10B), F major (7B), G major (9B), A minor (8A), E minor (9A).
- Audio payload requests
- 0
- External requests during workflow
- 0
Scope limit: Six clean triad progressions do not establish accuracy for modal, atonal, noisy, sparse or changing-key recordings. Treat every real-song result as an estimate.
Read this suite's machine-readable cases →
Question: Can the UI measure controlled signals, normalize them to -23 and -16 LUFS, re-measure the render and export a readable PCM16 WAV?
Observed: Both measurement, normalization and download workflows passed: -23 LUFS target → -23 LUFS in the UI, 384,044-byte WAV; -16 LUFS target → -16 LUFS in the UI, 384,044-byte WAV.
- Audio payload requests
- 0
- External requests during workflow
- 0
Scope limit: The fixtures are steady mono 1 kHz tones and the independent cross-check is not a certified broadcast meter. This wave covers sample peak, not true peak.
Read this suite's machine-readable cases →
Question: Can the real UI turn a stereo 48 kHz WAV into a mono 22.05 kHz float32 WAV while preserving duration and both authored tones at the expected downmix levels?
Observed: The two-second stereo fixture became a 22,050 Hz mono float32 WAV (176,444 bytes); both authored tones remained at the expected downmix levels.
- Audio payload requests
- 0
- External requests during workflow
- 0
Scope limit: This case covers WAV-to-WAV conversion, one downmix and one output rate. MP3 encoding and every browser-decodable input codec remain outside this wave.
Read this suite's machine-readable cases →
Question: Can the real waveform handles keep exactly 1.00–3.00 seconds of a four-segment fixture and exclude the authored outer tone segments from the downloaded WAV?
Observed: The visible handles selected 1.00–3.00 seconds and exported an exact 2.00-second PCM16 WAV (192,044 bytes) containing the 330/440 Hz middle segments with no measured 220/660 Hz leakage.
- Audio payload requests
- 0
- External requests during workflow
- 0
Scope limit: This case covers keep-selection with zero fades and pointer-position handles. Remove mode, fades and touch-specific gestures remain outside this wave.
Read this suite's machine-readable cases →
Question: Can the real UI reorder two decoded files, apply a 100 ms crossfade and export the expected sequence, duration and AudioContext-rate WAV?
Observed: After the visible reorder and 100 ms crossfade, the 1.7-second WAV followed the 660→330 Hz sequence at 48,000 Hz (163,244 bytes).
- Audio payload requests
- 0
- External requests during workflow
- 0
Scope limit: This case covers two mono inputs, one reorder and one crossfade. Chromium decodes both source rates to its AudioContext rate before the joiner sees them; larger queues, channel expansion and clipping control remain outside this wave.
Read this suite's machine-readable cases →
Question: Can the real UI route a controlled centred vocal and anti-phase side signal into instrumental and acapella WAV outputs without invoking the optional neural model?
Observed: Both instant-DSP exports passed: instrumental vocal/side ratio 0.028881, acapella side/vocal ratio 0.000002, with no neural-model request.
- Audio payload requests
- 0
- External requests during workflow
- 0
Scope limit: These cases cover the instant phase-DSP engine on one synthetic stereo mix. They do not test the optional neural engine or establish separation quality for mastered songs.
Read this suite's machine-readable cases →
Question: Can the real UI export bass, vocals, drums and other WAV stems with the expected authored routing and a sample-level sum close to the input fixture?
Observed: All four PCM16 WAV stems exported and summed back to the fixture with RMSE 0.00003146 and peak error 0.00009155; authored bass, vocal and transient/side routing checks passed.
- Audio payload requests
- 0
- External requests during workflow
- 0
Scope limit: The splitter is a frequency/transient heuristic, not an AI source-separation model. The fixture verifies deterministic routing and reconstruction, not perceptual isolation on real music.
Read this suite's machine-readable cases →
Question: Can the real UI learn a deterministic stationary noise floor, reduce broadband noise and hum, retain the authored speech-proxy tones and export a readable WAV?
Observed: Default-strength cleanup reduced noise-only RMS by 40.39 dB and 120 Hz hum by 101.37 dB, retained 94.57% of the authored speech-proxy tones and improved SNR by 23.06 dB.
- Audio payload requests
- 0
- External requests during workflow
- 0
Scope limit: This case covers one seeded stationary noise profile and synthetic tones. It does not establish quality for changing noise, reverberation, overlapping voices or one-off sounds.
Read this suite's machine-readable cases →