SoundBakery

AI-powered audio sample manager for macOS — index, search, and browse your local sound libraries with natural language.

🔍

Find sounds by describing them: "warm vinyl crackle" or "dark cinematic riser" — no more digging through folders.

📈

Everything runs locally via MLX. No data leaves your machine. No internet required once models are downloaded.

📁

Point at any folder of audio files. SoundBakery scans, tags with AI, and builds a searchable catalog — WAV, MP3, AIFF, FLAC, and more.

📈

Audio files are automatically tagged by AI — acoustic descriptors, sound sources, and genre classifications.

🎶

Preview clips with waveform visualization, real-time effects (EQ, reverb, delay, pitch shift, and more).

📤

Export clips as WAV or MP3 with configurable quality, sample rate, and channel mixing.

Quick Start

Launch SoundBakery
On first launch you will see the Setup Wizard. Choose In-App AI (fully local) or Remote Provider (connect to an OpenAI-compatible server).
Download a model
If using local AI, download an LLM (Qwen3-1.7B recommended) and an embedder (all-MiniLM-L6-v2). Models are downloaded from HuggingFace — allow a few minutes.
Add a folder
Click Add Folder in the sidebar and select a directory of audio files. SoundBakery will scan, tag, and index everything automatically.
Search for sounds
Type a description into the chat or use the catalog browser to filter by duration, tags, or filename.
Preview and export
Click any result to play it with waveform visualization. Apply effects in real-time, then export to WAV or MP3.

SoundBakery main window — folders sidebar, catalog browser, and chat assistant

Remote API support — Use OpenAI, Groq, LM Studio, or any OpenAI-compatible backend for LLM inference.
15 audio effects — Pitch shift, EQ, reverb, delay, transient shaper, bitcrusher, and more.
10 languages — Full UI localization including Russian, Spanish, French, German, Japanese, Turkish, and more.
sqlite-vec search — Fast approximate nearest-neighbor vector search for instant semantic retrieval.
Apple Foundation tagger — Built-in Apple SoundAnalysis as an alternative to the bundled CNN model.