🎹 ISLA v3.1.0 — DSP Stability & Hear the Difference

Let the Music Flow. With Greater Precision.

ISLA is ALIVE. ISLA is LIVE. Now, ISLA is getting stronger.

Following the release of ISLA v3.0.0 — ALIVE + LIVE, version 3.1.0 introduces an important engineering milestone: improved numerical robustness, reproducible musical listening comparisons, and a more intuitive macOS audio setup experience.

This release focuses on making ISLA's existing audio-processing foundation more reliable while preserving its creative identity.

The mission remains unchanged: explore how mathematics can transform musical continuity, harmonic texture, and expression.

✨ What's New in v3.1.0

🔬 1. More Reliable Real-Time Audio Processing

Two reproduced numerical issues have been addressed in ISLA's experimental LIVE engine.

Improved startup reconstruction

The weighted overlap-add (WOLA) processor now correctly accounts for the normalization weights of virtual silent frames preceding the first incoming audio block.

Previously, incomplete startup normalization could create significant unintended amplitude amplification.

In controlled Melt-mode testing with a half-scale constant input:

Sample rate | Before | After -- | -- | -- 44.1 kHz | 1.238757 | 0.458413 48 kHz | 1.135527 | 0.458413

This correction removes the reproduced startup amplification in the tested conditions without changing ISLA's established processing delay, harmonic strengths, or mode architecture.

Improved spectral reconstruction

LIVE now applies a continuous, precision-aware spectral correction when FFT coefficients approach numerical nulls.

The updated reconstruction reduces sensitivity to small representation-scale perturbations without introducing arbitrary phase-dependent energy at exact spectral nulls.

Controlled numerical sensitivity decreased from approximately 10–12% to 0.063–0.087% relative RMS difference in the selected tests.

These measurements represent improved numerical stability, not an independently established improvement in perceived sound quality.

🎧 2. Hear the Difference — ISLA's First Listening Showcase

For the first time, ISLA provides a reproducible listening showcase directly from its GitHub repository.

Eight FLAC audio samples are included:

Four original LIVE baseline recordings: Bypass, Velvet, Legato, and Melt.
Four corresponding recordings using the numerical robustness corrections.

The showcase uses Emotional Piano by triangelx, distributed under the CC0 public-domain dedication.

The comparisons use the same six-second musical excerpt, consistent presentation edits, documented loudness matching, and lossless FLAC storage of 24-bit PCM audio.

Processing provenance, source attribution, gain adjustments, rendering details, and SHA-256 checksums are preserved.

Explore the listening comparisons:

Hear the Difference — Baseline Samples

Before & After — Numerical Robustness

Listeners can now explore ISLA's harmonic transformations without configuring a live system-audio capture session.

🎛️ 3. Original Musical Profiles Preserved

ISLA continues to offer three LIVE harmonic-processing profiles:

Velvet (0.35) — A gentle, softened harmonic texture.

Legato (0.55) — An exploration of greater musical connectedness.

Melt (0.75) — A more pronounced, flowing harmonic transformation.

Bypass — The original signal, preserved for comparison.

The controlled DSP investigation confirmed that the current LIVE profiles represent different strengths of one underlying harmonic transformation.

Version 3.1.0 intentionally preserves those settings.

Independent musical-mode redesign and additional artistic tuning remain subjects of future research.

🖥️ 4. Improved macOS and BlackHole Setup

The guided setup audio workflow now explains the distinction between selecting an output inside ISLA and configuring macOS's system-wide audio destination.

The required LIVE route is:

Mac system audio → BlackHole 2ch → ISLA LIVE → Physical speakers/headphones

Setup now more clearly instructs users to:

Select BlackHole 2ch as ISLA's capture input.
Select physical speakers or headphones as ISLA's playback destination.
Manually choose BlackHole 2ch under macOS System Settings → Sound → Output.
Start LIVE monitoring only after the system-audio route is configured.

If the system output is incorrect, the routing error now identifies the selected macOS output when available and explains the required correction.

ISLA continues to reject unsupported, ambiguous, or potentially unsafe audio routes. No automatic macOS routing changes have been introduced.

🧪 5. Expanded Regression and Research Evidence

This release incorporates a comprehensive controlled DSP investigation and additional regression tests.

Research and testing include:

Reproduction of numerical defects before applying corrections.
Independent overlap-add reconstruction verification.
Spectral-null and phase-stability experiments.
44.1 and 48 kHz testing in mono and stereo.
Exact delayed Bypass checks.
Warm processing-bank and bounded-state validation.
Same-source waveform comparisons and numerical analysis.
Synthetic CPU-load and worker-budget measurements.
Audio provenance, source checksums, and reproducibility instructions.
Cross-platform Python acceptance and package-build verification.

The completed DSP study examined 224 matched processing cases and 25,200 measured worker hops in its performance investigation.

No modeled worker-budget overruns were observed in that benchmark. Native audio callbacks, actual hardware latency, and physical-device reliability were outside its scope.

LIVE worker-test scheduling behavior has also received additional regression coverage, although intermittent simulated-worker timing failures observed in some CI runs warrant continued investigation.

📦 6. Version Alignment and Compatibility

ISLA's version metadata is now aligned at 3.1.0 across:

Python distribution metadata.
Public and legacy Python version identifiers.
HTTP service metadata.
JUCE/CMake project configuration.
Automated version regression tests.

Existing public interfaces, historical compatibility aliases, plugin identifiers, and DSP operating modes are preserved.

The original v3.0.0 release remains separately available as the historical ALIVE + LIVE milestone.

⚠️ Experimental Status and Known Limitations

ISLA remains an experimental audio research project.

LIVE currently targets macOS with BlackHole 2ch and an independently selected physical stereo output.

The Python-based monitoring engine is not guaranteed to meet hard-real-time audio requirements. Longer-duration monitoring, physical disconnection recovery, acoustic latency, broader hardware compatibility, and new controlled listening acceptance remain open engineering activities.

The existing offline nonzero-strength chunk-streaming expected failure also remains documented.

The JUCE plugin pathway requires separate native acceptance and should not be considered a certified commercial audio plugin.

No claim is made that v3.1.0 introduces three independently designed musical effects or guarantees perceptually superior sound.

🚀 Getting Started

Repository: https://github.com/dfeen87/ISLA

For the terminal listening experience:

python -m pip install -e '.[playback]'
python -m isla

At the isla> prompt, enter:

setup audio

For LIVE system-audio processing, install BlackHole 2ch separately and follow the README's macOS routing instructions.

For ALIVE local-file processing, BlackHole is not required.

🔭 Looking Ahead

The next stage of ISLA research will explore whether Velvet, Legato, and Melt can develop genuinely independent musical characteristics.

Areas of investigation include frequency-aware harmonic shaping, carefully bounded temporal smoothing, musical-envelope processing, listening-based evaluation, and further native audio reliability improvements.

Each experimental direction will be measured, tested, and evaluated before adoption.

The objective is not merely to make audio different. It is to investigate what makes a musical transformation meaningful.

❤️ Dedicated to Isla

ISLA is dedicated to Isla, the creator's niece.

What began as a personal exploration of musical transitions and mathematical signal processing continues to grow into an open-source audio research project.

ISLA's development reflects the contributions of human creativity, computational research, open-source collaboration, and AI-assisted engineering.

Thank you to everyone who has contributed to ISLA's ongoing evolution.