H14 — Recorder

The mobile recorder is the tunes-ios contribution client. It records locally, helps the passenger align the journey, and uploads only reviewed and consented derived data by default.

For: contributors, product designers, researchers, and tunes-ios implementers.

Assumptions: iPhone-first; offline-first during collection; station-to-station sections are the comparison unit; raw PCM stays on the device by default.

Recording workflow

flowchart TD
  A[Select route prior] --> B[Record locally]
  B --> C[Process and infer sections locally]
  C --> D[Review and edit timeline]
  D --> E[Trim or delete]
  E --> F[Add optional context]
  F --> G[Privacy and field preview]
  G --> H[Optional perception survey]
  H --> I[Explicit consent and upload]

Nothing leaves the device before preview and consent.

Before and during recording

Select system, line, direction, origin, destination, and branch when topology requires it.
Keep the phone in hand with the microphone unobstructed; avoid handling it during measurement.
Record audio and available motion data locally.
Continue collection without network access.
Preserve detected clipping, interruption, audio-route changes, and sensor gaps as quality evidence for later review.

The casual path should not require exact rolling-stock, train-unit, or carriage numbers.

After recording, the recommended coarse carriage selector is front / middle / rear of train. Optional context may include position within the carriage (end / centre / near doors / near bogie) and posture (sitting / standing).

These are passenger metadata, not professional measurement positions. They help interpret variation but do not upgrade acoustic quality.

Future work

The repository does not define a literal seat-map selector or exact seat identifiers. Any seat selector requires user testing and a generic representation that does not hard-code one train interior.

GPS and station alignment

GPS is optional evidence, not the primary underground clock. The recorder combines:

user-selected route and network topology
motion and acceleration/braking cues
dwell and acoustic-envelope changes
door/chime event hints
optional GPS where accuracy is useful

The preferred result is a snapped section ID and relative timeline, not a high-rate location trail. Timetables are soft priors, not ground truth.

Editing timeline

The post-recording editor follows a subtitle/chapter model:

Show a local level envelope and proposed from→to section blocks.
Allow boundaries to be dragged, stretched, split, merged, or reassigned within legal topology.
Mark dwell, held-between-stations, walking/interchange, partial, gap, and trimmed intervals.
Recalculate duration T when boundaries change.
Require explicit confirmation before public-comparison eligibility.

Ambiguous intervals should be excluded or flagged, never silently filled.

Offline mode and upload pipeline

Recording, local processing, alignment, editing, and consent review must work without connectivity. A consented derived package may be queued until a connection is available.

After upload, tunes-web validates the schema and consent metadata, applies quality gates, associates confirmed sections, and produces aggregates. Periodic versioned open releases are owned by tunes.

Future work

Retry rules, queue persistence, intake authentication, idempotency, withdrawal API behaviour, and server error contracts are not specified in this repository.

Metadata

Collected or derived for the default scientific path

Group	Examples
Journey	Network profile, line, direction, section IDs, duration, confirmed boundaries
Acoustics	Separate level, peak, and band summaries
Provenance	Device model, OS, app and processing versions, calibration state
Quality	Tier, clipping/dropout/processing/placement flags, confidence dimensions
Consent	Timestamp, consent class, upload-preview acknowledgement, fields sent, withdrawal handle
Optional context	Phone placement, train third, crowding, posture, carriage position
Optional perception	Structured responses, separate from acoustic metrics; instrument versioning is planned

Raw audio and detailed sensor traces may be collected locally to support processing, but are not part of the default upload.

Intentionally not required or not published by default

raw waveform audio
contributor identity or stable public personal identifiers
precise GPS trails
exact train/unit or carriage identity in casual mode
exact occupancy percentages
medical diagnoses or hearing-health claims
identifying free text
announcement transcripts or a public speech archive

Any future research-restricted raw or excerpt path requires a separate protocol, explicit consent, legal and ethics review, access controls, and retention rules. It is not a casual toggle.

How it works · Privacy · Quality tiers · Privacy and research ethics · Journey alignment · Passenger metadata · Data-flow boundaries · Raw-audio policy · Derived-only default · Subjective timing · Derived upload schema