CodeWithWinton/macbook-unibody-acoustic-tap-dataset

🎙️ MacBook Unibody Acoustic Kinetic Tap Dataset (TLM 1.5)

Dataset Description

This dataset contains 13,127 uniform 48.0kHz 500ms floating-point NumPy (.npy) and HDF5 (.h5) audio windows capturing physical kinetic impulse waves traveling across metallic MacBook aluminum unibody chassis.

It was engineered for MORSE, a software-defined acoustic AI engine powered by TLM 1.5 (Tap Learning Model).

• Repository: https://github.com/CodeWithWinton/morse
• Model Engine: TLM 1.5 (Tap Learning Model Engine)
• Authors: Manas Maheshwari (@CodeWithWinton) & Daksh Sethi
• License: Dual-Licensed (AGPL-3.0 / Commercial OEM License)

---

📊 Dataset Structure

• double_left_palm/ (3,000 raw .npy samples): Double-taps on the left metal palm rest.
• double_right_palm/ (3,000 raw .npy samples): Double-taps on the right metal palm rest (30cm away from built-in mic across aluminum deck).
• noise_and_typing/ (7,127 raw .npy samples): Multi-surface ambient noise, typing clacks, desk bumps, car cabin rumbles, Instagram Reels, and speech.

---

🔬 Physical Feature Highlights

• 3,730D Spatial Feature Matrix: Combines a 3,720 STFT Mel Spectrogram grid ($20 \text{ Mel Bins} \times 186 \text{ Time Frames}$) with 10 physical kinetic scalar features (spectral_tilt, spatial_hf_decay, onset_attack_slope, high_mel_skew).
• Position-Invariant Peak Alignment: Centers kinetic impact peaks at Sample 4,800 (100ms into 500ms window) for 100% streaming buffer alignment.
• Solid-State Aluminum Wave Dispersion: Exploits high-frequency ($2.5\text{kHz}-4.5\text{kHz}$) acoustic attenuation across 30cm of aluminum unibody metal.

---

🏆 Benchmark Performance

• 5-Fold Stratified Cross-Validation: 98.4% (+/- 0.1% SD) Mean CV Accuracy across 25,127 feature vectors.
• Unseen Test Set Evaluation (5,026 clips): 0.99 Weighted F1-Score and 100% Left Tap Recall.
• Live Rejection Precision: 99.8% Precision with zero false triggers during active typing.

---

📜 Citation

@misc{maheshwari2026morse,
  author = {Manas Maheshwari and Daksh Sethi},
  title = {MORSE: Software-Defined Acoustic Kinetic Impulse Sensing via Solid-State Unibody Wave Dispersion},
  year = {2026},
  publisher = {Hugging Face / GitHub},
  howpublished = {\url{https://github.com/CodeWithWinton/morse}}
}