How to Do Real Technology Enhances the Arts Research at Home: A High School Student’s Guide to Free Tools, Affordable Kits, and Public Databases

Tech-art research used to live inside film studios, music conservatories, and university HCI labs. Now a teenager with a laptop, a webcam, and a soldering iron can run the same kinds of experiments those labs ran ten years ago.

This guide is your starting point. It covers the three things you need to do serious Technology Enhances the Arts (TECA) research from your bedroom: the kit you can buy for hobby money, the free creative and AI software you can install today, and the public art, music, and motion datasets that count as real research data.

Why this is possible now

Generative models that used to need a rack of GPUs now run on a single consumer graphics card or a free Google Colab notebook. You can fine-tune a music model, synthesize a voice, or generate temporally coherent video from your room.

Sensors that used to cost thousands have collapsed to hobby prices. A used Kinect costs around $40, a Quest 2 headset is cheap on the secondhand market, and IMUs, EMG bands, flex sensors, and contact mics all sit under $30 each. Your phone is already a high-resolution camera, a microphone array, and a 6-axis IMU.

Creative software went free and professional at the same time. Blender, Godot, Unreal, Unity Personal, TouchDesigner non-commercial, Pure Data, SuperCollider, Krita, and Audacity are all production-grade and zero dollars.

Put those three together and a kitchen counter plus a laptop becomes a small studio that can record motion, render 3D, generate sound, train models, and run a controlled user study.

The Technology Enhances the Arts home kit

Group your purchases by what they do. You do not need everything below. Pick the subgroup that matches the project shape you want.

Compute and microcontrollers

• Laptop with a consumer GPU (or a free Google Colab / Kaggle account for training).
• Raspberry Pi 4 or 5 for real-time vision, audio, and projection projects (~$45 to $80).
• ESP32 or ESP32-S3 for fast sensor capture, BLE, and small ML inference (~$8 to $15).
• Arduino Uno or Nano for simple actuator and sensor control (~$10 to $25).

Cameras, displays, and projection

• Pi Camera or USB webcam for face, hand, pose, and gesture tracking (~$15 to $40).
• Mini DLP pico projector for projection mapping and structured light (~$80 to $150).
• Smartphone screen as a display, a Pepper's-ghost source, or a parallax-barrier base.
• E-ink Waveshare panel for slow, ambient generative art (~$30 to $60).
• WS2812B LED strip or 16x16 matrix, MAX7219 modules for light-based displays (~$10 to $25).

Motion, gesture, and biosignal sensors

• MPU-6050 or BNO055 IMU for hand, head, and wearable motion (~$5 to $20).
• MyoWare EMG sensor for muscle-driven interfaces (~$40).
• Flex sensors for finger and joint bending (~$8 each).
• MPR121 capacitive touch board for invisible touch surfaces (~$8).
• MAX30102 pulse and SpO2 sensor for heart-rate-driven art (~$5).
• Used Kinect v2 or Leap Motion for full-body and hand tracking (~$40 each).

Audio capture and output

• INMP441 I2S microphone for clean digital audio capture (~$5).
• Contact mic or piezo disc for vibration and instrument-body sound (~$3 to $15).
• Basic MIDI keyboard for music-tech projects (~$40 to $80).
• Cheap USB audio interface for instrument recording (~$30 to $60).

Actuators and tactile output

• SG90 or MG90 hobby servos for kinetic sculpture and puppetry (~$3 each).
• ERM or LRA haptic motors for vibration vests, belts, and gloves (~$2 to $8).
• Stepper motor + driver for gantries, turntables, and POV displays (~$10 to $25).
• Hobby galvanometer laser kit for vector laser drawing (~$60 to $100).

Fabrication

• FDM 3D printer with PLA or PETG for enclosures, sculptures, and lenses (~$180 to $300).
• Resin printer when you need fine surface detail (~$200 to $300).
• Laser-cut acrylic ordered online for diffraction gratings, parallax barriers, and lenticular mounts.

A complete starter kit, picked from the lists above based on your project, lands somewhere between $150 and $500.

Signature technique: real-time vision and gesture on a webcam

The one workflow that unlocks more TECA projects than anything else is real-time computer vision and pose tracking on a regular webcam using MediaPipe and OpenCV. Face, hand, body, and gaze become inputs you can wire into music, graphics, robots, and displays.

Here is the 5-step workflow.

1. Install MediaPipe, OpenCV, and Python. A clean conda or venv environment with mediapipe, opencv-python, and numpy is enough to start. Confirm your webcam streams at 30 frames per second.
2. Pick one signal stream. Choose hand landmarks, face mesh, body pose, or gaze. Run the demo script and read the output: 3D landmark coordinates per frame.
3. Smooth and map. Apply a one-Euro filter or simple exponential smoothing to the landmarks. Map a useful derived value (fingertip distance, head yaw, mouth openness) to a parameter (volume, hue, servo angle, particle speed).
4. Send the signal somewhere creative. Push the mapped value over OSC into TouchDesigner, Pure Data, or SuperCollider. Or send it over MIDI into your DAW. Or send serial into an Arduino driving servos and LEDs.
5. Run a small user study. Have 10 to 20 people use your interface against a baseline (mouse, keyboard, no smoothing). Collect a Likert rating, a task-completion time, and short interview notes.

That single loop, sensor to model to creative output to user evaluation, is the spine of an ISEF-grade TECA project.

The dry-lab side: free software you can install today

Generative AI for images, audio, and video

• Stable Diffusion for image generation, in-painting, and ControlNet-guided art.
• ComfyUI for visual node-graph control of diffusion pipelines.
• Hugging Face Diffusers for scriptable image, video, and audio generation in Python.
• AnimateDiff and SDXL-Lightning for short generative video clips on a single GPU.
• MusicGen, AudioLDM, and Riffusion for text-to-music and text-to-sound.
• Bark and Piper for text-to-speech in many voices and languages.
• RVC and Coqui-TTS for voice conversion and custom voice cloning.
• Whisper for high-quality speech transcription and translation.
• Demucs and Spleeter for separating vocals, drums, bass, and other from a mix.
• BasicPitch and CREPE for polyphonic and monophonic pitch tracking.

Graphics, 3D, and fabrication

• Blender with Python scripting for modeling, animation, and procedural geometry.
• Three.js and Babylon.js for browser-based 3D scenes you can demo on any laptop.
• NeRFStudio, Instant-NGP, and Gaussian-Splatting for photogrammetry and novel-view synthesis on a consumer GPU.
• OpenSCAD and FreeCAD for parametric, code-driven 3D models.
• MeshLab and Open3D for point cloud and mesh cleanup.
• scikit-image marching cubes for turning math equations and volumes into printable surfaces.

Creative coding and live media

• Processing and p5.js for sketch-based generative visuals.
• OpenFrameworks for high-performance C++ creative coding.
• TouchDesigner non-commercial for real-time visuals driven by sensors, OSC, and MIDI.
• Pure Data, SuperCollider, ChucK, FAUST, and Sonic Pi for programmable sound and synthesis.
• Audacity and Reaper free trial for recording, editing, and mixing.

Game engines

• Godot for open-source 2D and 3D games.
• Unity Personal and Unreal Engine for higher-fidelity 3D and VR builds.

HCI, sensing, and ML

• MediaPipe, OpenPose, and dlib for face, hand, body, and gaze tracking.
• PyTorch and torchaudio for training custom models on small datasets.
• librosa for audio feature extraction.
• spaCy, sentence-transformers, and Stanza for text and language analysis.
• Ollama and llama.cpp for running small local LLMs (Mistral, LLaMA family) on your own machine.

Running the same software a working artist or researcher would use changes the way the project feels. You are not making a "school version", you are running the real pipeline.

Public databases that count as real data

Music and audio

• MAESTRO for paired MIDI and audio piano performances.
• NSynth for tens of thousands of labeled instrument note recordings.
• FMA and MTG-Jamendo for large licensed music libraries with tags.
• MagnaTagATune and GTZAN for classic music tagging and genre tasks.
• AudioSet and VGGSound for general-purpose labeled sound events.
• URMP and MIDI Lakh for multi-track classical and aligned MIDI data.

Speech and voice

• LibriSpeech and CommonVoice for transcribed speech in many languages.
• RAVDESS and IEMOCAP for emotion-labeled speech.

Images and art

• WikiArt, BAM!, and Behance for labeled art and illustration.
• LAION-Art and ArtBench-10 for art-specific image-text pairs.
• Quick-Draw and Sketchy for sketch and stroke data.
• COCO for general object and scene labels.
• Met Open Access, Rijksmuseum, and Google Arts & Culture API for high-resolution museum artwork.

Motion, gesture, and sign

• How2Sign, WLASL, MS-ASL, ASL-LEX, and SignBank for sign language video and linguistic data.
• MUSIC-AVQA for paired music performance audio and video.

You are allowed to do real research entirely on top of these. Re-analyzing public data with a new method, a new evaluation, or a new cross-modal pairing is itself a legitimate research path.

How to combine wet and dry: the strongest project shape

Pattern A: sensor in, model in the middle, creative output out. Capture a real signal from a real person (gesture, breath, voice, EMG, gaze) on hobby hardware. Run it through a model or DSP pipeline you wrote. Drive a creative output (sound, light, motion, image). Evaluate with a controlled user study.

Pattern B: public dataset in, model in the middle, hardware artifact out. Train or fine-tune on a public art, music, or motion dataset. Take the trained model out of the notebook and into a physical artifact (a 3D-printed sculpture, a projection display, a wearable, a kinetic installation). Evaluate against a baseline with objective metrics and listener or viewer ratings.

Judges respond strongly to this hybrid shape because it shows you can do both the engineering and the human evaluation, which is exactly what real TECA research demands.

Choosing a phenomenon that has not been done

1. Search Google Scholar for the two or three core nouns of your idea (for example, "Pepper's ghost head tracking" or "raga constrained transformer"). Read the abstracts of the top 15 results. Note what they did and what they did not measure.
2. Search the Society for Science abstracts archive (the public ISEF and Regeneron STS abstract database) for your category and your keywords. You are checking the shape of recent student work, not copying it.
3. Search the ACM Digital Library, IEEE Xplore, and arXiv (under cs.HC, cs.SD, cs.GR, and cs.CV) for the most recent two years of work on your technique. Filter by year, sort by relevance.

If you find nearby work, that is good news, not bad news. Adjacent prior work proves your problem is real and gives you baselines to compare against.

A realistic timeline

• 1 to 2 weeks: replicate one published pipeline (for example, MediaPipe hand tracking into a Pure Data synth) and measure one variable.
• 1 to 2 months: full hybrid project with a clear baseline, a built artifact, and a user study with 15 to 30 participants. Strong shape for a regional fair.
• Full year: ISEF-track project with a novel method, multiple ablations, a real evaluation protocol, and a written paper. Includes a pilot study, a revision, and a final study.

If this is your first research project, start with the 1-to-2-week version. You will learn what is hard before you commit a year to it.

A starter checklist

1. A clean, lit workspace with room for a tripod, a webcam, and a small build area.
2. A free Google Colab account and a Hugging Face account.
3. A local Python environment (conda or venv) with numpy, opencv-python, mediapipe, librosa, torch, and torchaudio installed.
4. Blender and one creative-coding environment (TouchDesigner non-commercial, Processing, or Pure Data) installed and tested.
5. A digital lab notebook (a dated Markdown file or a Notion page) where every session gets a timestamped entry.
6. Your microcontroller toolchain installed (Arduino IDE or PlatformIO for ESP32 / Pi setup).
7. A written one-line research question of the form "Does X change Y compared to baseline Z, measured by metric M?"

Once those seven items are checked, you are ready to pick a phenomenon.

Where to go next

• Display Technology (DSP): new ways to show images, including volumetric displays, projection mapping, lenticular and parallax-barrier 3D, and dynamic light-based displays.
• Human Information Exchange (HIE): translation, captioning, sign language, tactile communication, and assistive interfaces between people and across languages.
• Music and Image Manipulation (MIM): generation, restoration, classification, and expressive control of music, audio, and images.
• Games (GAM): new input modalities, accessibility-driven controls, adaptive difficulty, generative levels, and AI-driven non-player characters.
• 3D Modeling (MOD): generative and computational pipelines for creating, printing, and customizing 3D objects.
• Engineering Effects (ENG): physical, kinetic, and robotic systems built for performance, theater, and live shows.
• Other (OTH): cross-modal art (scent, taste, haptics), biometric-driven generative art, museum interaction, and provenance for AI art.

Each subcategory has its own MehtA+ project guide built around the kit on this page. Pick the subcategory you are most curious about and start there. The studio you used to need a school, a stage, or a software license to enter now fits on your desk.

Project ideas in this category (70)