The science behind NITE DSP.
We build creative tools around actual audio signals. By doing math on your local CPU cores rather than sending files to remote servers, we guarantee absolute privacy and zero workflow latency.
The Local Extraction Pipeline
How SLO reads, analyses, and understands audio signals without relying on file names.
Windowed Spectral Analysis
Raw PCM audio buffers are windowed using Hann functions and processed via standard Fast Fourier Transforms (FFT) at 2048-sample window lengths, establishing high-resolution frequency domain mapping.
Mel-Frequency Spacing
Power spectra are mapped onto the Mel scale, reflecting the logarithmic pitch spacing of human hearing and extracting 13 mel-frequency cepstral coefficients (MFCCs) to map timbral character.
Transient Signature Extraction
Sub-band energy changes are tracked across frames to identify transient onset rise times and decay envelopes, determining the structural envelopes of sounds.
Acoustic Dimension Reduction
MFCC vectors and envelope parameters are combined into high-dimensional signatures. Local dimension reduction models map these down to a coordinates grid, placing similar sounds near each other.
Built for local performance.
C++ Vectorised Similarity Checks
The similarity engine uses hardware-accelerated dot product instructions (using Apple Silicon NEON registers and Intel AVX instruction sets) to check up to 10,000 files in under 2 milliseconds.
Local SQLite Index Architecture
A local database is written to ~/Library/Application Support/NITE/SLO/. Instead of altering your original library directory structure, the virtual browser indexes files safely in read-only mode.
Zero Cloud Telemetry
We believe your creative library is your competitive advantage. By running 100% offline signal classification in memory, your original works remain on your hard drives and never touch external servers.