Voilib: Open Source Podcast Search Engine 🔍

Voilib offers semantic search in thousands of minutes of high-quality transcriptions of podcasts. Just type your query and it will find related content in thousands of episodes. Voilib also allows users to index their own audio files.

🎧 Try it now for free at voilib.com!

Voilib doesn't offer any paid service. Help me ensure the continued availability and accessibility of it by supporting me through donations. You will directly contribute to covering the server expenses and enable me to maintain and improve Voilib for users worldwide. Customized assistance is also offered.

▶️ run your own instance now!

You can run your own instance of Voilib in your server, it doesn't depend on any external paid service.

mkdir voilib && cd "voilib"
curl https://raw.githubusercontent.com/unmonoqueteclea/voilib/main/compose.yml -o compose.yml
docker compose up

You will need an admin user and password. By default user voilib-admin with password *audio*search*engine will be created.

You can change default ports with environment variables:

• VOILIB_MANAGEMENT_PORT (for management page: default 8501)
• VOILIB_FRONTEND_PORT (for frontend: default 80)
• VOILIB_API_PORT (for backend: default 81)

After all services are up, jump to http://localhost:8501 and follow the instructions to populate Voilib with content. You can also check first run tasks section.

More information about deployments in infra/readme.

❓ how it works

Voilib performs 4 main tasks: collecting, transcribing, indexing and querying podcasts episodes to find the most interesting fragments for every user prompt.

• collection: Almost all public podcasts have an associated RSS feed that contains metadata about every episode and a link to the audio file. Voilib uses those feeds to collect and store that metadata from the list of podcasts configured by the application admin. Additionally, Voilib can also index your own audio files.

• transcription: The collected episodes are then transcribed using Whisper: Open AI's Open Source Transcription Model.

• index: Episodes transcripts are divided into fragments of approximately 40 words (check DEFAULT_FRAGMENT_WORDS constant to see the value currently used). Then, Voilib calculates the embedding of each fragment. In that way, every fragment is converted into a vector of 384 floating point numbers (check EMBEDDINGS_SIZE constant to see the embedding size currently used). Those vectors are stored in a vector database: Qdrant.

• queries: For each new user prompt, Voilib just needs to calculate the embedding of it and find the closest ones in the vector database, returning the most relevant episodes fragments to the user.

license

Voilib is licensed under the GNU GPLv3 license. See COPYING.

Permissions of this strong copyleft license are conditioned on making available complete source code of licensed works and modifications, which include larger works using a licensed work, under the same license. Copyright and license notices must be preserved. Contributors provide an express grant of patent rights.