Voice Keeper — Final Project Masterpiece

Voice Keeper is an offline audio playback station for physical printed photographs — a synthesis of Fab Academy skills from Weeks 01–19. Place a photo on the tray, touch the play sensor, and the device reads a colour-encoded strip on the photo edge and plays the linked MP3 from a microSD card. No screen, no cloud, no recording UI — only paper, sound, and a domestic object on the shelf.

“Every photo has a voice, every voice tells a story.”

Fab Academy final project requirements

This page documents the masterpiece required by Fab Academy 2026 — integrating the range of units covered and answering each checklist question below. The project is independently operable (runs without a laptop once programmed) and demonstrates individual mastery of design, fabrication, and programming.

• ✓ Integrated design — multiple digital fabrication processes in one product
• ✓ 2D & 3D modelling applied to my own designs
• ✓ Additive (FDM) and subtractive (laser cut, PCB mill) fabrication
• ✓ Custom fabbed microcontroller PCB with input & output
• ✓ System integration and packaging
• ✓ Summary slide and ~one-minute video (conception → construction → operation)
• ✓ Make rather than buy where practical (enclosure, PCB, firmware, encoding workflow)

What Does It Do?

Voice Keeper links printed photographs to audio memories. Each photo carries a 1 cm quinary (base-5) colour strip on its bottom edge — four blocks in Black / Blue / Green / Red / White, encoding an ID from 0 to 624 (5⁴ combinations). The user slides the photo into the front tray so the strip covers four sensor windows, then touches the play sensor (TTP223B). The XIAO ESP32-C3 reads four TCS34725 colour sensors via a TCA9548A I2C multiplexer, computes the ID, and commands a DFPlayer Mini to play the matching xxx.mp3 through an 8 Ω speaker.

A simplified Web UI on the ESP32 (adapted from Week 15) can trigger playback by ID for setup and testing. The product stores photos inside the box (≈120+ capacity) and audio on the microSD card. Target users: new parents, families preserving elders’ voices, and anyone who prefers physical albums over phone galleries.

Who Has Done What Beforehand?

Voice Keeper builds on existing ideas but combines them differently:

• Talking photo frames / RFID albums — link images to audio via hidden tags; Voice Keeper uses a visible colour strip on the print (human-readable, no chip embedded in the photo). Early ideation (January) considered RFID; March–April evolved to colour encoding for cost and transparency.
• Digital picture frames with audio — usually screen- and cloud-centric; Voice Keeper is offline and tactile.
• Colour-as-data projects — multicolour encoding in maker art; this project uses a defined quinary alphabet sized for TCS34725 sensors and family-scale albums (~100 photos, 625 ID capacity).
• Fab Academy weekly prototypes — Week 8 PCB, Week 11 sensor/audio tests, Week 15 interface patterns, and Week 16 integration de-risked the final merge.

Full prior-art, applications, and dissemination discussion: Week 18 — Applications and Implications, Week 19 — Invention, IP and Income.

What Did I Design?

• Product concept — Voice Keeper philosophy, interaction flow, target users (see Product Philosophy below).
• Encoding system — 4 × base-5 colour blocks; read the ID as a four-digit number:"C4 C3 C2 C1"; every digit is from 0 to 4, strip layout template for printed photos.
• 2D mechanical — laser-cut plywood panels (Laser Maker + DXF: box_design_v2.dxf).
• 3D mechanical — FDM rotary top cover, colour-sensor window frame (Top_cover v2.stl, color_sensors_frame.stl); skills from Week 02 and Week 05.
• Electronics — custom milled PCB from Week 08 (XIAO ESP32-C3, button D0, LEDs D1/D2); carrier for I2C, UART, power in the integrated unit.
• Embedded firmware — TCS34725 polling via TCA9548A, quinary decode, DFPlayer UART control, touch input, optional Web UI (subsystems tested in Week 11; merged in final loop).
• Content workflow — assign ID → record MP3 (e.g. TTSMaker) → print photo with encoded strip → store in box.
• Documentation — this GitLab Pages site; photo prep via SmallerPic; wildcard fabric×3D-print experiments in Week 17.

What Sources Did I Use?

• Fab Academy 2026 curriculum, regional reviews, and Chaihuo instructor feedback
• Component datasheets: TCS34725, TCA9548A, DFPlayer Mini, Seeed XIAO ESP32-C3, TTP223B
• Adafruit / DFRobot libraries (TCS34725, NeoPixel where used)
• Chaihuo lab equipment: laser cutter, PCB milling machine, 3D printer, multimeter
• Fab community tools: Gerber2PNG, Mods PCB mill workflow (Week 08)
• Peer documentation and group assignments
• My weekly assignment logs — indexed at the bottom of this page

Materials, Components, Sources, and Cost

What Parts and Systems Were Made?

Made in the Fab lab (individual mastery):

• Milled custom PCB (Week 08)
• Laser-cut plywood enclosure and internal structure (Week 03)
• 3D-printed rotary cover and colour-sensor frame (Week 05)
• Embedded firmware — sensor, audio, touch, Web UI subsystems
• Colour calibration workflow and encoded photo strips
• Internal wiring harness and integrated assembly (Week 16)
• Documentation website and SmallerPic tool (SmallerPic)

Bought modules (integrated and documented, not used as opaque black boxes):

• TCS34725 boards, TCA9548A, DFPlayer Mini, speaker, XIAO ESP32-C3, touch sensor, battery/charge boards

Integrated systems: optical ID (4× sensor + mux), audio playback (UART + SD), human input (touch sensor), power (Li-ion + USB), mechanical packaging (laser + print + paint). Full assembly plan in Week 16.

What Processes Were Used?

Voice Keeper integrates the full Fab Academy range. Each row links to weekly documentation:

Note: Week 12 (group vending machine) and Week 14 (molding/casting of VK logo) are documented separately; they informed mechanical thinking but are not core Voice Keeper subsystems.

Demonstrating Fab Academy Competencies

Product Philosophy and Technical Architecture

Voice Keeper emerged from a simple observation: we store thousands of photos in cloud albums but lose the sounds tied to them — a grandparent’s voice, a child’s first words, ambient noise from a trip. The device is a physical vessel for emotional memory: tactile photos, invisible digital encoding, privacy-respecting offline operation.

Colour encoding: quinary (base-5) mechanical computing — Black(0), Blue(1), Green(2), Red(3), White(4). Four sensor positions → 5⁴ = 625 IDs. Strip: 1 cm at photo bottom edge, four 8 mm × 8 mm blocks.

Electronics block diagram:
XIAO ESP32-C3 → I2C → TCA9548A → 4× TCS34725 (colour ID); UART → DFPlayer Mini → speaker; GPIO → TTP223B touch (play); 5 V rail → logic + audio + sensors; Li-ion + USB-C charge.




Interaction: place photo in tray → touch play sensor → read colours → play MP3. Optional Web UI for ID-based playback during setup.

Early ideation sketches (January–March):


The second idea was to make a keypad to enter the ID for playback, but it was not implemented because I did not have enough time.
Neil provided a suggestion when I first time present my idea during the lecture. I actually tried the following version by using 8 light sensors.


As I decided to keep using XIAO, and there were not enough pins to connect 8 light sensors, I finally decided to use the color sensors instead.

System Integration

Week 16 documents how mechanical, electronic, and software subsystems come together — integration plan, CAD/sketches, packaging, and finished-product design.

Full documentation: Week 16 — System Integration

• Footprint ≈ 164 mm × 120 mm × 75 mm; laser-cut plywood + 3D-printed rotary cover
• Integration sequence: power → electronics tray → touch panel → sensor deck → speaker → closure
• Finished look: painted shell, hidden wiring, USB-C on back, photo tray with groove rail

What Questions Were Answered?

• Can a Fab lab student build a custom PCB that boots and runs peripherals? Yes — Week 8 board + Weeks 9–11 tests on the same pin map.
• Can TCS34725 sensors read distinct colours over I2C with a mux? Yes on the bench — stable values per swatch (Week 11); in-box calibration ongoing.
• Can DFPlayer play reliably from XIAO UART? Yes — 9600 baud, track trigger works (Week 11).
• Laser-cut plywood vs pure 3D print for enclosure? Plywood chosen for rapid prototyping and tray geometry; 3D print used where plywood failed (rotary cover).
• Can ESP32 serve a useful control UI? Yes — WS2812B web controller (Week 15) adapted for DFPlayer setup.
• How do subsystems fit in one box? Integration sequence and risks documented (Week 16).

Still open: colour reliability under ambient room light without extra LEDs; full calibration table for blue vs green; battery runtime under speaker load at volume 28/30. See Week 19.

What Worked? What Didn't?

Working:

• Integrated physical product — painted box, rotary cover, sensor deck, speaker, touch play
• Individual subsystems on bench — colour sensors, DFPlayer, touch sensor, Web UI
• Week 8 PCB as reusable platform through input/output/networking weeks
• Laser-cut + 3D-print hybrid packaging; plywood cover replaced by printed part when it broke
• Clear product story, encoding spec, and documentation pipeline (site + slide + video)

Not working yet / risky:

• Colour discrimination under warm indoor light — calibration incomplete for production reliability
• TCA9548A multiplexer needs more in-box testing under full assembly
• Speaker and touch sensor exposed in v1 — planned covers for v2
• Power under speaker load — bulk capacitance and routing need verification on integrated unit

How Was It Evaluated?

Evaluation criteria (from Week 18):

• Functional: demo photos — correct audio within ~2 s of touch play; Web UI triggers correct track by ID
• Technical: all subsystems present (sense, compute, play, input, power); firmware and PCB files in repo
• Integration: single enclosed unit; USB-C charge; microSD accessible for content updates
• Design: reads as finished product, not breadboard — painted shell, hidden cables
• Documentation: this page, weekly assignments, summary slide, ~60 s video, BOM, make-vs-buy explicit
• Individual mastery: I design, build, and program the unit myself; it runs standalone

What Are the Implications?

Applications: offline memory objects for families — grandparents hear voices without apps or accounts; new parents link lullabies to printed milestones.

Dissemination: open documentation on GitLab (CC BY-NC); final presentation slide and video; optional post-Fab workshop at Chaihuo. See Week 19 — dissemination plan.

Intellectual property: no patent planned; quinary strip encoding and design files shared openly for remix; third-party libraries credited.

Income (optional): custom commissions or maker workshops — realistic only after family beta proves colour-read reliability. See Week 18 and Week 19.

Broader implication: the Fab Academy skills chain (2D/3D → subtractive/additive fab → electronics → embed → integrate) turns a personal idea into a reproducible open-hardware artifact, not just a slide deck.

Summary Slide and One-Minute Video

Fab Academy requires a one-page summary slide and a ~one-minute video showing conception, construction, and operation.

Summary slide: presentation.png


Video (~60 s): presentation.mp4

Video storyboard: title + power on → place photo → touch play → audio → design sketches → laser cut / mill / solder / program / assemble / 3D print → paint & decorate → Web UI demo → final hero shot → power off.

Presentation schedule: Chaihuo — Final Presentation

Weekly Assignments — Contribution to Final Project

License and Repository

Documentation and design files: Creative Commons Attribution–NonCommercial (CC BY-NC). Source code, firmware sketches, SmallerPic, STL/DXF exports, and weekly assignments live in gitlab.fabcloud.org. Voice Keeper demonstrates individual mastery of Fab Academy skills integrated into one independently operable masterpiece.