Inspiration
A squeak is only a source. For a short horror scene, a creator still needs to decide where the unseen presence is, how it approaches, and when it is revealed. Foley Séance turns those choices into a small visual sound-design instrument.
What it does
Import a short sound or choose a synthetic starting sound. Place its spirit in a haunted chamber, choose or draw movement, set the door and an optional opening or closing event, and shape the echo. Render the cue, compare a revision against a retained reference, and download a stereo WAV. Save a local project to carry source and settings into another session.
The intended use is a preauthored cue for a short film, fixed-camera scene or cutscene. The spooky interface is functional: spirit position, path and door state govern audible transformations. That is why we chose Costume Contest.
How we built it
The static browser app uses JavaScript modules, Web Audio playback and worker-based rendering. It requires no account, microphone or cloud audio service. Export is stereo PCM16 WAV at 48 kHz. The source builds with Node.js; an optional Python loopback launcher serves the generated app. The public source is MIT licensed.
Kiro CLI supplied bounded drafts for room DSP, audition core/controller/worker components, an echo engine patch, and a motion normalization/retiming helper. Root .kiro/provenance preserves source-safe prompts, outputs and attribution. Generated drafts were reviewed, corrected and integrated rather than accepted unchanged. Later timed-door, reference-take and save-reminder work is integrator-authored. We do not claim a spec/hook/steering workflow we did not use. Runtime audio processing is deterministic, not generative AI inference.
Codex-assisted work integrated and reviewed components, developed the surrounding UI and project format, wrote regression tests, checked save/reopen and export paths, and iterated on layout and recovery. It also helped assemble the demo from actual application captures and Joseph's recorded narration.
Features
- Short local imports or synthetic sources.
- Chamber-based stereo placement, distance coloration, authored movement and sampled drawn paths.
- Initial door state plus one timed opening or closing event, with artistic echo.
- Explicit render, playback and WAV export.
- One tab-only reference take for sequential revision comparison.
- Local project download/reopen with source and settings and an unsaved-change reminder.
Testing and challenges
The source suite reports 259 source/model tests and four launcher checks. Hosted browser rendering, WAV export and reference playback were checked. Save/reopen and render-byte preservation were exercised in recorded browser checks. These qualify the tested paths, not every browser or artistic outcome.
The difficult part was keeping visual controls, saved project state and rendered audio consistent while making failure recovery understandable. We corrected generated fixtures, hardened input handling and separated transient audition/reference state from persisted project data.
Try it
Open https://josepha-mayo.github.io/foley-seance/ in a modern desktop browser and start at a low listening volume. Choose a built-in sound, select Stage a hidden presence, review and confirm the proposed changes, then Summon & listen. Download the WAV. Keep a reference, change echo or movement, render again and compare. Save a project, check the browser download, reload and reopen it.
Source: https://github.com/josepha-mayo/foley-seance
No credentials are needed. For local source use, run node build.mjs, then python3 run-local.py or the Windows launcher with existing Node.js and Python 3.9+.
Scope and limitations
This is fixed-perspective artistic processing, not physical acoustics or player-relative game audio. Inputs are averaged to mono; opposing channels can cancel. Compressed decoding is not strictly resource-bounded by the accepted file limits. Use short trusted clips. There is no autosave or general undo; confirm downloads before closing. Drawn motion samples relative timing and is not keyboard editable, although existing movement choices are keyboard accessible. Reference playback uses the same monitor gain, not matched perceived loudness. Current Windows end-to-end usability, broad touch support and human artistic quality remain unqualified. Creator time savings have not been measured.
Credits and disclosure
Built-in sounds are synthetic; the app bundles no third-party recordings. The demo uses the CC0 cabinet-squeak preview by joedeshon, Freesound 339184: https://freesound.org/people/joedeshon/sounds/339184/. Corridor imagery is an illustration, not game-engine integration. Five historical replay/adaptation harnesses have disclosed publication-copy path parameterization; the original Kiro outputs are unchanged. No novelty is claimed for stereo panning, echo or door automation.
Built With
- css
- html
- javascript
- kiro-cli
- node.js
- python
- web-audio-api
- web-workers