Paranormal Investigation Command Center
Beyond The Horizon Labs — BTHLCorp
Features • Build • Usage • Whisper AI • Hardware • User Guide
BTHL-SpiritBox is a multi-sensor paranormal audio detection and transcription system. It integrates SDR frequency sweeping with real-time audio demodulation, AI-powered voice detection, electromagnetic field correlation, and comprehensive session archival.
The spirit box technique rapidly sweeps radio frequencies, and our system automatically detects voice-like patterns in the demodulated audio, transcribes them using local AI (whisper.cpp), and correlates them with EMF sensor data to identify potential paranormal phenomena.
Companion app to BTHL-SpectraSentry Tactical Skanner.
SDR (SoapySDR) → IQ Samples → Audio Demodulator (AM/FM/SSB)
├→ Speaker Output (PortAudio)
├→ Voice Activity Detector → Whisper AI Transcription
├→ EMF Correlator ← GQ EMF-390 Serial Reader
└→ Session Recorder (WAV + JSON + CSV)
| Feature | Description |
|---|---|
| SDR Frequency Sweep | SoapySDR multi-device support with configurable profiles |
| Audio Demodulation | AM, NFM, WFM, USB, LSB modes with DC removal and decimation |
| Voice Activity Detection | Multi-feature analysis: RMS energy, zero-crossing rate, spectral flatness |
| AI Transcription | Local whisper.cpp speech-to-text (no cloud, no internet required) |
| EMF Monitoring | GQ EMF-390v2 serial protocol with spike detection |
| Temporal Correlation | 500ms sliding window matches EMF spikes with voice events |
| Session Recording | WAV audio + JSONL events + CSV EMF timeline + audio snippets |
| Widget | Description |
|---|---|
| Spectrum Waterfall | Scrolling spectrogram with BTHL cosmic gradient |
| Audio Waveform | Real-time waveform + VU meter + VAD confidence bar |
| EMF Timeline | Scrolling time-series with spike markers and value labels |
| Detection Log | Color-coded filterable event log (voice/transcription/correlation/EMF) |
| Profile | Range | Steps | Modulation |
|---|---|---|---|
| FM Broadcast | 88-108 MHz | 101 | WFM |
| AM Broadcast | 530-1700 kHz | 118 | AM |
| VHF Low Band | 30-88 MHz | 58 | NFM |
| Full Spectrum | All bands | Variable | Mixed |
brew install qt@6 soapysdr soapyrtlsdr portaudio pkg-config cmake# Clone whisper.cpp into lib/
git clone https://github.com/ggerganov/whisper.cpp.git lib/whisper.cpp
# Download a model
mkdir -p ~/BTHL-SpiritBox/models
curl -L -o ~/BTHL-SpiritBox/models/ggml-base.en.bin \
https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-base.en.binmkdir build && cd build
cmake .. -DCMAKE_BUILD_TYPE=Release -DCMAKE_PREFIX_PATH="/opt/homebrew"
cmake --build . -j$(sysctl -n hw.ncpu)./BTHL-SpiritBox- Launch — the app auto-detects your SDR and EMF-390
- Profile is set to FM Broadcast by default (88-108 MHz)
- Click START SWEEP — waterfall fills, audio plays
- Voice detections appear in the Detection Log with confidence scores
- EMF spikes correlate with voice events for high-confidence anomalies
- Click Load Model in the Transcription dock
- Browse to
~/BTHL-SpiritBox/models/ggml-base.en.bin - Transcriptions appear automatically when voice is detected
- Click Record — creates timestamped session directory
- All audio, events, and EMF data are archived
- Click Stop to end recording
- Sessions saved to
~/BTHL-SpiritBox/sessions/
- Voice (green) — VAD confidence triggers
- Transcription (blue) — Whisper AI text results
- Correlation (gold) — EMF + voice coincidence events
- EMF Spike (red) — Significant electromagnetic deviations
| Device | Purpose | Connection |
|---|---|---|
| RTL-SDR (via SoapySDR) | RF frequency sweeping | USB (any SoapySDR-compatible device) |
| GQ-EMF-390v2 | EMF/EF sensing | USB serial (CH340, VID 0x1A86) |
# Required for RTL-SDR
brew install soapyrtlsdr
# Optional for other SDR devices
brew install soapyhackrf soapyremoteSoapySDRUtil --findThe Voice Activity Detector uses three audio features with weighted scoring:
| Feature | Weight | Voice Characteristic |
|---|---|---|
| RMS Energy | 30% | Voice is louder than background noise |
| Zero-Crossing Rate | 30% | Voice: 0.02-0.15, noise: higher |
| Spectral Flatness | 40% | Voice has spectral peaks, noise is flat |
Confidence Formula:
confidence = 0.30 × energy + 0.30 × zcr + 0.40 × flatness
Triggers when confidence > 0.5 and duration > 100ms.
The EMF Correlator uses a 500ms sliding window to detect temporal coincidence between voice detection events and electromagnetic field spikes. Correlation score factors:
- Temporal proximity — how close in time the events occurred
- EMF spike strength — deviation from rolling baseline
- Combined confidence — weighted product of VAD and EMF scores
bthl-spiritbox/
├── src/
│ ├── core/ SweepEngine, AudioDemodulator, SweepProfile
│ ├── audio/ AudioOutputManager, AudioRingBuffer (lock-free SPSC)
│ ├── detection/ VoiceActivityDetector, WhisperTranscriber
│ ├── emf/ EMFSerialReader, EMFCorrelator
│ ├── hub/ SensorHub (multi-sensor orchestration)
│ ├── sensors/ SensorDevice, USBDeviceManager
│ ├── session/ SessionRecorder, SessionPlayer
│ ├── ui/ MainWindow + 4 visualization widgets
│ └── main.cpp
├── profiles/ Sweep profile JSON files
├── docs/ User guide, architecture documentation
├── lib/whisper.cpp/ whisper.cpp submodule (optional)
└── CMakeLists.txt
See docs/ARCHITECTURE.md for detailed module descriptions and thread model.
Each recording session creates a directory with:
sessions/2026-04-12_20-30-15/
├── session.json Session metadata (ID, timestamp, devices)
├── audio_sweep.wav Complete audio recording (16-bit PCM)
├── events.jsonl All detection events (newline-delimited JSON)
├── emf_timeline.csv EMF readings with spike flags
└── audio_snippets/
├── voice_001.wav Individual VAD audio clips
├── voice_002.wav
└── ...
| Document | Description |
|---|---|
| User Guide | Complete usage instructions and best practices |
| Architecture | System design, data flow, and module descriptions |
| CLAUDE.md | Development guide for Claude Code |
BTHL-SpiritBox is designed as a companion to BTHL-SpectraSentry. Both apps share the GQ EMF-390 and RTL-SDR hardware (one app at a time). Future integration via IPC bridge will enable cross-app anomaly correlation.
Copyright (c) 2026 BTHLCorp — Beyond The Horizon Labs. All rights reserved.
- David St. John — Project Lead, BTHLCorp
- Claude Opus 4.6 — Architecture & Code Generation
- Open source: Qt 6, SoapySDR, PortAudio, whisper.cpp, yaml-cpp