SFX Scout
AI-Powered Semantic Sound SearchDescribe what you hear in your head using natural language. Surface the perfect sound effect across your entire multi-database library in milliseconds. Offline, local, and built for professionals.
Core Pillars
Semantic Search
Type exactly what you're hearing: "sharp high-frequency blade whoosh" or "heavy underwater explosion." LAION's CLAP neural net maps sound semantics, ignoring simplistic text-matching limits.
Audio Similarity
Surfaces sonic matches in a click. Select a sound, hit Similar, and FAISS-backed IndexHNSW shards will instantly pull related catalog variations, discovering lost gems in seconds.
100% Local & Private
Your library is your competitive edge. SFX Scout processes everything entirely on your hardware. Zero data leaves your machine, zero subscription fees, and no cloud latency.
System Features
CLAP Neural Querying
Translate natural language directly into matched audio files. SFX Scout maps descriptive keywords to sonic patterns, bypasses text limits, and delivers precise spatial matches.
FTS-Backed Fallback
Direct filename, category, and catalog searches match instantly. If no exact term is hit, the system falls back to semantic vector models without dropping performance.
Fielded Search Queries
Target specific metadata fields directly from the main search bar using commands like library:TRex Bev or source:My Library for quick scoping.
Paged Search Results
When result limits are selected, you can navigate through paginated result groups instead of being strictly hard-capped at the initial search limit.
Full Metadata Import
Supports over 22+ metadata fields from Soundminer imports including Keywords, Designer, Category, CD Title, and Show names without losing structural fidelity.
Smart Folder Relinking
Relink missing files easily. Select rows in your table, point the database to a directory tree, and SFX Scout will recursively scan and relink records in place.
Standalone Directory Scanning
Add normal folders. The built-in scanner infers category, subcategory, and descriptions from folder organization and file-naming structures automatically.
Compaction & Audits
Convert older databases, normalise catalogs, prune unreferenced data, and export failed embedding lists or duplicate rows directly to CSV for review.
In-Process JUCE Host
Native audio I/O, output-device channels, mono sum, processing chains, and plugin editor windows run directly inside one shared C++ JUCE library in the app process.
Single-Instance Rack Plugins
Instantiate VST3/AU slots inside the app. Slot engines stay alive across bypasses, keeping plugin meters, processing states, and native editor windows unified.
Multi-Algorithm Engine
Pitch shift and time stretch using Signalsmith (smooth), Rubber Band (precise transients), or librosa. Features Paulstretch for extreme 8x-100x+ ambient stretches.
Spectrogram Analysis
See spectral content as you preview. Blend a frequency spectrogram over the interactive waveform display using a smooth, blendable opacity slider.
Browse & Random Modes
Explore freely. Use Browse mode to see catalog folders, lock and search strictly within active results, or pull random sets of sounds with a click.
Concat & Spot Rendering
Render temporary concatenative WAV variations. Spot-to-timeline features render active pitch, speed, and plugin layers so spotting output matches what you preview.
Persistent Theme Toggling
Toggle between Dark and Light UI layouts in a single keypress (Alt+T). Window geometry, custom database color codes, and table widths persist across sessions.
Shortcuts Master
Total key control. Toggles looping (L), fine-adjusts varispeed speed (Shift++/-), resets DSP (/), shuffle results, and navigates without using a mouse.
CLAP-Large Embeddings
Leverages 512-dimensional vector models from LAION's high-fidelity CLAP models to capture complex cross-modal sound relations and audio meanings.
Dual Indexing Strategy
Maximizes performance. FAISS HNSW shards handle audio-to-audio similarity searches, while IVF shards process text searches with sub-500ms latency.
Metal Performance Shaders
Fully optimized for Apple Silicon hardware. Uses native macOS Metal (MPS) on M-series chips to process huge databases locally at maximum speed.
Idle Resource Releasing
Smart RAM footprinting. FAISS index shards and heavy embedding models unload when idle for a period, keeping background resource usage minimal.
engineered for cinematic sound post-production.
Brings advanced AI sound description maps to your local drive. No hidden subscriptions, zero cloud latency, and complete pipeline control.
| Feature Matrix | SFX Scout | Competitor A | Competitor B | Competitor C | Competitor D |
|---|---|---|---|---|---|
| AI Semantic Search | ✅ CLAP Neural | ❌ Keyword Only | ⚠️ Thesaurus | ❌ Keyword Only | ✅ Vector AI |
| Audio Similarity | ✅ FAISS HNSW | ❌ No Match | ❌ No Match | ❌ No Match | ✅ Standard |
| Pricing Tier | TBA | $199 - $899 | $150 / yr | Free - $649 | Free / Licensed |
| Data Privacy | ✅ 100% Local | ✅ 100% Local | ❌ Cloud DB | ✅ 100% Local | ✅ 100% Local |
| In-Process DSP Host | ✅ JUCE Plugins | ✅ Radium Mixer | ⚠️ Basic Pitch | ✅ Standard DSP | ❌ No Pitch |
System Specs & Setup
# Clone the source repository
$git clone https://github.com/cjguitar/sfx-semantic-search.git
$cd sfx-semantic-search
# Create local virtual environment and activate
$python3.11 -m venv .venv
$source .venv/bin/activate
# Install requirements and boot client UI
$pip install -r requirements.txt
$python -m src.gui.app