Sonoteca V2: semantic retrieval for institutional sound archives

A web architecture built for Universidad de las Artes (UArtes) bridging natural-language descriptions and institutional audio collections through multilingual embeddings and vector search.

The challenge

Moving beyond brittle filenames and rigid tag taxonomies

University sound archives and post-production libraries frequently rely on cryptic recorder filenames or inconsistent manual tags. Students and sound designers looking for an acoustic space, emotional tone, or foley texture rarely know an asset's exact ID.

Sonoteca V2 addresses this bottleneck by enabling descriptive natural-language queries centered on narrative intent, acoustic environment, and sonic character.

FastAPIPython web backend
1024-dimmultilingual-e5-large embeddings
QdrantVector similarity database
UArtesInstitutional archive preservation
System architecture

Semantic audio retrieval pipeline

The platform pairs multilingual dense text embeddings with fast vector similarity indexing to surface relevant recordings for film students, lecturers, and researchers.

01 · QueryNatural-language prompt (EN / ES)
02 · Embeddingsentence-transformers + multilingual-e5-large
03 · IndexQdrant vector similarity ranking
04 · AuditionOrdered audio results for creative review
Python / FastAPIsentence-transformersmultilingual-e5-largeQdrantSemantic Audio RetrievalHuman–Computer Interaction
University context

UArtes teaching & preservation

  • Built for Music and Sound Production students, lecturers, and researchers at Universidad de las Artes (Guayaquil).
  • Turns institutional field recordings and sound collections into discoverable teaching and production resources.
  • Integrated into studio coursework on sound design, film mixing, and creative AI.
Production workflow

Designed for working ears

  • Bridges the gap between a sound editor's narrative vocabulary and archive retrieval.
  • Supports bilingual querying across Spanish and English inside one shared embedding space.
  • Augments creative curation without replacing critical human listening.
Research ecosystem

UPV PhD & studio tooling

  • Aligned with doctoral research in Art: Production and Research at Universitat Politècnica de València (UPV).
  • Complements peer-reviewed work in Fuera de Campo (Vol. 8, No. 2) on AI documentary synchronization alongside the audio tool AlineaFase.

Research & semantic archives

Collaborate on Sonoteca V2

[email protected]