Quinlight Audio is a tracker music player and remastering tool for MOD/S3M/XM/IT and related formats. It plays modules, can remaster their source samples with optional external AI backends (AudioSR, LavaSR, FLowHigh, AP-BWE), and lets you A/B the result live during playback.
- Supported public target:
x86_64-unknown-linux-gnu - Supported posture: Linux-first public release, not a cross-platform build
- Default playback/export target: 96 kHz, 32-bit float (64-bit mixed end-to-end)
- Plays tracker formats through vendored libopenmpt with a double-precision mixer
- Opens modules directly from archives (
.zip,.7z,.rar,.tar.*,.lha,.cab,.iso) - Replaces samples live during playback so you can compare Original, Reference 48k, and AI remasters (AudioSR, LavaSR, FLowHigh, AP-BWE) without restarting the song
- Combines multiple AI engines into a single sample via a per-bin spectral consensus on the rotor manifold — bins where the engines agree pass through, bins where they disagree (the typical hallucination fingerprint) get auto-suppressed
- Exports the live result to FLAC or AAC (256 kbps)
- Supports batch CLI rendering for directories of modules
- Installs as a Linux desktop app (
--install-icon)
Quinlight Audio works without AI engines installed. The player, archive support, reference cleanup path, and export flow remain available even if you never set up the optional remaster backends.
Quinlight Audio's vendored libopenmpt fork is rebuilt for end-to-end double-precision audio. Every stage from sample interpolation through mixing to output uses 64-bit floating point — the only quantization in the playback path is the final cast to f32 at the audio device.
The entire mixer bus operates in double (mixsample_t = double). Volume,
panning, interpolation, and filter feedback all accumulate in 64-bit precision.
Volume ramps use Hermite smoothstep curves (t²(3−2t)) instead of linear ramps,
eliminating zipper artifacts on note transitions. The channel filter is a
cascaded 4-pole design — IT-style 2-pole resonant biquad followed by a
Butterworth post-filter — for 24 dB/octave rolloff with no integer truncation
in the coefficient path.
Each sample in the module can be upscaled to 48 kHz via three methods:
- AI (AudioSR / LavaSR / FLowHigh / AP-BWE): neural bandwidth extension
- 48k reference: deterministic sinc resampling (FFmpeg swresample)
- Original: raw sample at native rate (typically 8–22 kHz)
Samples are replaced live during playback. Pattern offset effects (Oxx, SAx)
are automatically rescaled to match the new sample rate, and portamento effects
are compensated in the engine so pitch slides sound identical regardless of which
sample mode is active.
Each enabled AI engine produces its own 48 kHz remaster of every sample. Quinlight Audio scores each candidate against the original by Pearson correlation of magnitude spectra below the source's Nyquist (an engine that hallucinates even at known frequencies isn't to be trusted), then combines the engines that pass via a per-bin Karcher mean on the rotor manifold ℝ⁺ × S¹:
- Magnitude — geometric mean of the engine magnitudes (Karcher mean on ℝ⁺ under multiplication). Smoothly biased toward the quieter engines: the rotor-correct successor to softmin, without the per-bin discrete-winner ringing of patched-together spectra.
- Phase — circular mean of the engine phases (Karcher mean on S¹).
- Agreement scaling — the resultant length of the phase rotor sum (0–1) multiplies the consensus magnitude. Bins where the engines agree on phase pass through at full amplitude; bins where they disagree (the typical hallucination fingerprint) are attenuated proportionally.
Below the source's original Nyquist the consensus is then rotor-blended back toward the source spectrum itself (arithmetic-mean magnitude, shortest-arc SLERP on phase) so the bottom band stays anchored to the ground truth and the engines contribute mainly to the band-extension above. Above the source Nyquist the consensus passes through unchanged.
Why operate on the rotor manifold instead of a Cartesian (complex-linear) blend: the chord between two phasors of comparable magnitude in ℂ passes closer to the origin than either endpoint when their phases disagree, so a linear blend silently attenuates the bin in proportion to phase mismatch — which the inverse STFT renders as pre-echo and transient smearing. Operating on the geodesic of (ℝ⁺ × S¹) makes that attenuation explicit instead of hidden: phase agreement modulates magnitude on purpose, which both sounds cleaner and is interpretable as a hallucination-rejection criterion rather than a silent artifact.
Pitch bends (vibrato, portamento, slides) are tracked in full double precision
(PitchT = double, FreqT = double) — no fixed-point period tables or integer
slide accumulators. IT linear slides use pow(2.0, amount/768.0) directly.
The resampling filter is a 64-tap polyphase sinc with 65536 phases (16-bit phase resolution) and an octave-spaced mipmap chain. Each mipmap level tunes Kaiser window beta independently (β = 14.0 at unity down to β = 8.0 at 128× downsample) with anisotropic velocity shear coefficients (k_β = 0.65, k_β² = 0.15) that widen the transition band in proportion to playback speed, keeping the stopband clean during fast pitch sweeps.
Full derivation and design notes: audio_anisotropic_filter_v2.pdf.
SIMD kernels are compiled for SSE2, AVX, AVX2, and AVX-512 with fully unrolled accumulator loops — runtime dispatch picks the widest available path.
A/B ten tracker modules straight from the repo. The before column is the deterministic render (original samples, no AI); the after column is the same module with samples upscaled by the AI engines. Both clips are 48 kHz MP3 at 320 kbps — downsampled from the engine's 96 kHz default and served via GitHub Pages so every browser plays them inline. Click to listen.
| Module | Format | Before | After |
|---|---|---|---|
| 2ND_PM | S3M | listen | listen |
| 4mat_-_eternity | XM | listen | listen |
| beyond_the_network | IT | listen | listen |
| Caroline | XM | listen | listen |
| GroovyUntightFunk | XM | listen | listen |
| jt_mind | XM | listen | listen |
| jt_pools | XM | listen | listen |
| sweetdre | XM | listen | listen |
| tiny_tunes | MOD | listen | listen |
| znm-wopeace | IT | listen | listen |
Same 20 clips at the engine's native 96 kHz / AAC 256 kbps, also served via GitHub Pages. Many browser AAC decoders accept 96 kHz files but resample to the system output rate at playback time — what you hear may not be the full 96 kHz, but the bytes your browser fetched are.
| Module | Before | After |
|---|---|---|
| 2ND_PM | listen | listen |
| 4mat_-_eternity | listen | listen |
| beyond_the_network | listen | listen |
| Caroline | listen | listen |
| GroovyUntightFunk | listen | listen |
| jt_mind | listen | listen |
| jt_pools | listen | listen |
| sweetdre | listen | listen |
| tiny_tunes | listen | listen |
| znm-wopeace | listen | listen |
Prefer the bundled download? Grab
quinlight-audio-96khz-bundle.zip
(275 MB — same 20 clips organized as rendered/ and remastered/).
Quinlight Audio currently targets Linux x86_64-unknown-linux-gnu. The build expects
Rust, a C++ toolchain, SDL2 headers, libarchive headers, and FFmpeg development
libraries.
Disk space: Plan for at least 30 GB free before installing. The full footprint (build artifacts + Python venv + AI model checkpoints) lands around 26 GB, with headroom for caches and rendered output.
sudo apt install build-essential clang mold libsdl2-dev libarchive-dev \
libavcodec-dev libavformat-dev libavutil-dev libswresample-dev libswscale-dev
cargo build --releaseThe supported public install path is the checked-in Linux installer:
./install_prerequisites.shThat script creates ~/.local/share/quinlight-audio/venv, installs the pinned Python
package set used by Quinlight Audio, and runs a simple smoke check at the end.
Supported AI matrix for this release:
- Platform: Linux
x86_64-unknown-linux-gnu - Python:
3.12+ - PyTorch:
2.11.x - TorchAudio:
2.11.x - TorchVision:
0.26.x
The GUI shows the same pinned commands if the engines are missing.
# Launch the GUI
quinlight-audio
# Launch with GPU remastering
quinlight-audio --upscale-mode gpu
# Render a module to FLAC or AAC at the default 96 kHz target
quinlight-audio render track.s3m -o track.flac
quinlight-audio render track.s3m -o track.aac --format aac
# Batch render a directory
quinlight-audio convert mods -o renders --format flac aac
# Restrict to specific engine(s)
quinlight-audio convert mods -o renders --engine audiosr --engine lavasr --engine apbwe
# Skip AI remastering (render originals only)
quinlight-audio convert mods -o renders --no-remaster
# Reference-only cleanup output (no AI, just cleaned 48kHz reference)
quinlight-audio convert mods -o renders --reference-only --cleanup-preset declick-ar
# Open modules from archives
quinlight-audio render mods.zip -o track.flac
quinlight-audio render mods.zip --file track.s3m -o out.flac
# Install .desktop file and icon
quinlight-audio --install-iconQuinlight Audio is built by Kind Computers, LLC. If it's useful to you and you'd like to help fund continued development, you can sponsor the project on GitHub:
❤ Sponsor Quinlight Audio on GitHub
AI backend redistribution and branded promotion should still be reviewed engine-by-engine before any bundled or company-branded release. This repository documents a supported external-install flow for those backends; it does not claim that backend weights are bundled or cleared for redistribution.
Patent pending. Quinlight Audio's multi-engine AI consensus algorithm — the per-bin Karcher-mean spectral consensus on the rotor manifold described under Multi-engine consensus — is the subject of a pending U.S. patent application.
MIT