Text to audio
Describe what you need in natural language and land on timed moments inside long files.
Audio Search
Find audio by description or sample.
Status: Available by request
Search for matching moments in your recordings. Audio Search runs on your computer. Audio, queries, and the search index stay on disk; nothing is uploaded.
- Runs on
- Your computer
- What leaves
- Nothing. The embedding model is downloaded once; audio and queries stay on disk.
- Access
- Available by request; email [email protected]
For archives and sound libraries.
Find a passage in a sensitive recording or a sound in a professional collection without uploading either library.
Private archive search
For recordings that need to stay on your computer, including investigations and internal voice archives.
- Query by sample or natural language without leaving the machine
- Moment-level hits inside long recordings
- Collections and CSV/JSON export for case or matter files
Audio professionals
Search sound effects, field recordings, production sound, podcasts, and radio archives locally.
- Trim a sample, find similar passages, and listen to compare results
- Watched folders with chunk windows and overlap you control
- History, pinned searches, Favorites, and export into your pipeline
Search audio the way you search documents.
Organize libraries, watch folders for new files, and compare results at specific moments in a recording.
Audio to audio
Drop a sample—or a trimmed region—and find similar passages across your indexed libraries.
A local search index
The model turns audio and text into numerical representations for matching. This processing runs on your device, with optional GPU acceleration.
Libraries & watch folders
Keep the index current as files appear. Choose the length and overlap of audio segments to search.
Compare, collect, and export
Compare hits, pin favorites into project bins, and export CSV or JSON when work leaves the app.
Command-line tools
Automate indexing and search for ops pipelines that prefer a terminal.
Where your data goes.
The embedding model downloads once. After that, indexing and search run locally. Your audio, text queries, embeddings, and results stay on disk. No cloud embedding service receives them.
- Chunked indexing with temporal grouping of hits
- Similarity ranking to find candidate matches for review
- Index backup and restore
- Built for private archives and professional sound libraries
Use it to find recordings worth listening to. A similarity score is not proof of what was said or who said it. Review the source audio before drawing conclusions.
The same local-embedding approach is the retrieval half of the private RAG systems we build.
Search your audio library locally.
Audio Search is available by request. Tell us about your recordings and the searches you need to run.