REFERENCE
Synthetic voice · matched level
0:00
SYNTHETIC HUMANITIES LAB / EXPERIMENT 01
One poem. Different deliveries.
What can the existing amplitude-based fly receiver distinguish?
← The listening chamberLoading the recorded comparison…
Frozen MaleCNS / fly.ai. Identical reset and seed set. No antenna model, training, or semantic input. Seeds sample simulator noise; they are not different flies.
COMPARISON
0:00
Playing one recording pauses the other. Both neural traces are ensemble summaries; the moving cursor follows the selected recording. The acoustic plots show the processed audio used by the simulator.
Navigate to the same source line
Select a line to seek both recordings to its own timestamp. No time stretching. Human timing is approximate; reversed-order diagnostics retain original line identities.
RESPONSE / what changed?
Paired difference = (comparison − its silence) − (reference − its silence). SD describes paired differences across seeds. These descriptive checks do not establish statistical significance.
Neural trajectories
— Reference · — Comparison · shaded bands: observed seed minimum–maximum
Elapsed seconds, 100 ms bins, Hz per neuron relative to matched silence. Dashed vertical lines mark audio endings; one second of aftermath is recorded. No extrapolation between unequal durations.
Paired difference through time
Supplementary descriptive trace metric, added after inspecting the initial mean-rate results; not a significance test.
Only complete 100 ms bins within the shorter audio duration are compared here. Positive means a larger rate change in the comparison, not preference. Curves are aligned by elapsed time, not by words.
What entered the receiver?
Processed waveform · signed sample range per 20 ms
Injected drive · 20 ms RMS mapping
Same waveform RMS does not guarantee equal mean frame RMS, capped drive, or total exposure. Duration and integrated injected drive are reported above. The original gain and cap are unchanged.
Exploratory annotated population differences
Non-input annotated types with at least ten neurons, ranked by absolute mean paired rate difference. This selection is exploratory; direction agreement is not a corrected significance test.
| Population | Neurons | Mean Δ | SD | Signs + / − / 0 |
|---|
Individual seeds and exact controls
| Seed | Reference − silence | Comparison − silence | Paired Δ |
|---|
READING / interpretation of the experiment
This short account is generated by transparent rules from paired response statistics alone. It receives no poem text. “More activity” does not establish enjoyment, fear, or a performed action.
Method and downloadable evidence
These are new level-matched experiments. The listening chamber retains its earlier single-seed recordings. The two should not be confused.
The original inputs here are the previously published PCM files; their prior normalization and source provenance are retained. Diagnostic variants alter synthetic line segments or verified digital silence. All simulations retain the original connectome and model parameters.
Full spike arrays are retained in the local experiment archive. The public count archive contains per-step global/group counts and phase counts for every annotated cell type, including silence controls. The manifest records local raw-file hashes.