Traction substitute recording — selected by the user on 2026-10-22

Title: 北京地铁15号线 VVVF驱动音.ogg
Author: BIT0865
Recorded: 2020-10-24, own work
Description by the uploader: 北京地铁15号线 ALSTOM OPTONIX IGBT-VVVF
Source: https://commons.wikimedia.org/wiki/File:北京地铁15号线_VVVF驱动音.ogg
License: Creative Commons Attribution-ShareAlike 4.0 International
https://creativecommons.org/licenses/by-sa/4.0/

This is explicitly a SUBSTITUTE recording. It is NOT Nanjing Metro Line 2
original audio and does not establish that the traction equipment is identical.

Edits from the approximately 55.12-second original, made 2026-10-22:
- traction-start.wav: source 0.0–18.4 s, startup excerpt.
- traction-roll.wav (revised): source 18.0–22.6 s, 0.7 s equal-power circular
  crossfade, producing a 3.9 s loop. The former 23–29 s passage had strong
  bass-level surges and is excluded. A smooth 55 Hz sub-bass reduction and
  offline circular RMS-envelope balancing reduce recurring volume pulses.
  The waveform boundary is matched; the loop ends are not faded to silence.
- traction-brake.wav: source 30.0–51.0 s, actual braking excerpt, not reversed
  acceleration. The appropriate final portion is selected for the remaining
  station approach time, preserving original pitch and playback speed.
All excerpts converted to mono, 22050 Hz, 16-bit PCM; DC offset removed;
short boundary fades on one-shot excerpts and conservative peak scaling applied.
The continuous rolling loop has no fade-to-silence at its boundary.

Player processing: continuous rolling bed, adaptive startup-to-rolling
crossfade, rolling-to-braking crossfade, final fade on stopping. Unity playback
rate preserves pitch. Recordings are RMS-balanced and pass through a low shelf,
slow-release peak limiter (no fast bus compressor). Pausing/seeking selects the matching excerpt
position; the rolling loop remains one continuous source. Delayed UI frames do
not restart the recording when its playback clock is still in sync.
These derived audio files are licensed CC BY-SA 4.0.

Nanjing du-du-du opening/closing cue remains separately credited in CREDITS.txt.
Device synthesized speech remains optional and off by default.

2026-10-23 phase balance revision:
Startup: smooth offline gain shaping makes its initially quiet first seconds
audible; a broad 800 Hz presence lift and static transient soft limiting retain
the original timing/pitch. Rolling: broad presence EQ plus circular envelope
balancing improves audibility without live AGC. Braking begins at the same
deceleration window as the visual speed model, with reduced foreground gain
and a continuing wheel/rolling bed. No new sound source was introduced.
The station-to-moving transition explicitly starts the traction excerpt; an
earlier transition bug could leave only the quiet rolling bed. A final soft
peak guard catches exceptional transients at maximum user volume, without
a fast gain-control loop on steady running sound.

2026-10-24 continuous mix revision:
Acceleration, steady running and braking gains now follow the current physical
speed every animation frame, fixing a stale departure-level gain during cruise.
Braking retains a continuous rolling bed, with lower foreground brake gain.
The long journey adds cruise time at unity audio playback rate. During supplied
announcements the extra traction bed makes room for speech; quiet extensions
restore it without restarting the loop. Tunnel coverage smoothly adds a low
shelf and three short filtered reflections (23, 41 and 67 ms, no feedback);
open-air output stays clearer. This is acoustic simulation, not a new Nanjing
recording. Existing original/derived audio file licenses remain unchanged.