v0.5.1 — Two-Pass NR, Chain Reorder & Tuned Defaults

What's New

14-Stage Vocal Pipeline (Phase 8e)

Reordered the processing chain and added a second noise reduction pass to fix the gain-rider-amplifies-noise problem.

New chain order:

1. Noise Gate → 2. NR (Pass 1) → 3. Gain Rider → 4. De-Plosive →
5. NR Cleanup (Pass 2) → 6. De-Reverb → 7. HPF → 8. EQ →
9. Comp Peak → 10. Comp Body → 11. De-Esser → 12. Soft Clipper →
13. Reverb → 14. Limiter

Key changes:

• Gate and NR now run before the gain rider, so noise is cleaned at its natural (quieter) level
• NR Cleanup (Pass 2) added after the gain rider — a gentle stationary-mode pass that catches residual noise the rider amplified, without stem-guided profile
• No more need for "Strong" NR to compensate for gain-rider-boosted noise — eliminates underwater/musical artifacts

Tuned Defaults

All effect defaults updated from real-world vocal testing:

• All 14 effects enabled by default
• Gain rider: max boost 4 dB (was 6)
• NR Cleanup: Strong
• HPF: 120 Hz (was 100)
• EQ: Clean Up preset (was Bright)
• De-esser: 5000 Hz / 12 dB (was 6000 / 6)
• Soft clipper: drive 1.8 (was 1.5)
• Reverb: enabled by default (0.15 wet, 25 ms predelay)
• Enhanced preset updated to match

Stats

• 14/14 effects working
• 232 tests (12 new), all passing
• v0.5.0 → v0.5.1: 5 files changed, +582 / -287 lines

v0.4.0 — De-Esser + Reverb (9/9 Effects Complete)

Phase 8c — Completes the 9-stage vocal enhancement pipeline

All 9 vocal effects are now fully implemented. No more stubs or "(coming soon)" labels.

New Effects

• De-Esser (chain step 7): Frequency-selective sibilance compressor with bandpass detection (4th-order Butterworth), envelope smoothing, dB threshold/gain computation, and attack/release envelope follower. Supports split-band mode (attenuates only the sibilant band, preserving vocal body) and wideband mode.
• Reverb (chain step 8): Dual-mode reverb — Schroeder algorithmic (4 parallel comb filters + 2 series allpass) or convolution via user-provided impulse response file (FFT-based). Includes predelay, wet/dry mix, and automatic fallback to algorithmic on IR load failure.

UI

• De-esser controls: frequency (3000–10000 Hz) and reduction (0–12 dB)
• Reverb controls: wet mix (0–0.50), predelay (0–80 ms), IR file browse button (WAV/FLAC)
• Enhanced preset now enables de-esser (split mode, 6 kHz, 6 dB reduction)
• Reverb remains off by default in all presets

Bug Fix

• Exported processed vocal now includes the alignment shift (auto-align or manual offset slider). Previously the exported file was unaligned.

Effects Pipeline (complete)

#	Effect	Status
1	Noise Gate	Working
2	Spectral NR	Working
3	De-Reverb	Working
4	High-Pass Filter	Working
5	Parametric EQ	Working
6	Compressor	Working
7	De-Esser	New
8	Reverb	New
9	Limiter	Working

Tests

203 passing (+14 new, −4 removed stub tests).

v0.3.1: NR Mode Selection + Advanced Parameters

What's New

Noise Reduction Mode Selection

• Auto mode (default): uses stationary algorithm when a reliable stem-guided noise profile is available, falls back to adaptive mode otherwise
• Stationary mode: best for constant room noise (fan, hiss, untreated room) with a good noise profile
• Adaptive mode: handles varying noise, less dependent on profile quality

Advanced NR Parameters Dialog

• Stationary threshold sensitivity (0.5–3.0)
• Frequency mask smoothing (100–2000 Hz)
• Temporal mask smoothing (10–200 ms)
• GPU (CUDA) toggle with auto-detection

Updated Defaults

• High-pass filter: 80 Hz → 100 Hz
• Parametric EQ default preset: Clean Up → Bright
• Limiter ceiling: -1.0 dB → -0.7 dB

Testing

• 10 new tests (7 noise reduction + 3 effects wrapper)
• 189 total tests passing

Full Changelog: v0.3.0...v0.3.1

v0.2.0 — Signal Cleanup, Presets & Mix UX

What's New

New Effects (5/9 working)

• Noise Gate — RMS-envelope gating with configurable threshold, attack/release/hold, and reduction depth. Silences inter-phrase gaps without clipping word onsets.
• De-Reverb — Spectral flux-based subtraction that reduces room reverb. Three strength levels (Light / Medium / Strong).

Preset System

• Raw — all effects off (passthrough)
• Clean — noise gate + noise reduction + HPF + limiter
• Enhanced — adds de-reverb on top of Clean

Presets bulk-configure all toggles and parameters. Manual changes auto-switch to "Custom".

Mix UX Improvements

• Auto-Tune Volume — automatically balances vocal and minus faders using LUFS measurement (vocal ~3 dB above minus)
• Per-track level meters — horizontal bars with green/yellow/red zones and peak hold, animate during playback
• Fader-based mixing — Mix & Export now reads track fader gains directly, so preview matches export
• Dynamic status labels — "Not yet processed:" updates to "Processed:" when effects are applied; same for mix result

Additional Features

• Effects chain bypass — global checkbox to skip all effects (for pre-enhanced vocal imports)
• Processed vocal export — save the V-proc result as a standalone WAV
• Jump-to-click sliders — all volume faders and seek slider jump directly to clicked position

Bug Fixes

• Mix result waveform and multi-track playback offset corrected (was off = 0, now off = max_lag)

Stats

• 162 tests passing
• 9 files changed, 823 insertions, 111 deletions

Install

pip install git+https://github.com/Artemarius/VocalForge.git@v0.2.0
# Or with Demucs separation support:
pip install "vocalforge[separation] @ git+https://github.com/Artemarius/VocalForge.git@v0.2.0"

VocalForge v0.1.0

First release of VocalForge — a desktop app for recording vocals over music tracks with automatic alignment, normalization, and mixing.

Features

• Song import — load any audio file as the original track
• Vocal separation — extract instrumental using Demucs (htdemucs_ft)
• Recording — play minus track while recording mic input
• Track preview — switch between Song, Minus, Vocal, and Mix Result; seek via waveform click or slider
• Automatic alignment — cross-correlation with constrained search window (background music or vocal stem matching)
• Noise reduction — stem-guided spectral gating + configurable high-pass filter
• LUFS normalization & mixing — ITU-R BS.1770-4 loudness normalization with adjustable vocal/instrumental balance
• Export — save final mix as WAV/FLAC

Install

git clone https://github.com/Artemarius/VocalForge.git
cd VocalForge
python -m venv venv
source venv/bin/activate
pip install .
vocalforge

For Demucs separation support: pip install ".[separation]"

Requirements

• Python 3.10+
• Audio interface with mic input
• GPU optional (recommended for faster separation)