FINCA International

Audio AI that transcribed, anonymized, and indexed thousands of hours of multilingual field interviews, making them queryable by meaning, not keywords.

Speech AINLPSemantic searchPython

6languages

4countries

Focus

Speech AI · Semantic search

Thousands of hours of field interviews across four countries and six languages were sitting unused in audio files. The system transcribed, anonymized, and indexed the multilingual audio, then made it queryable by meaning rather than keywords, taking regional teams from weeks of listening to finding answers in seconds.

Highlights

▹Multi-language transcription and speaker handling
▹PII anonymization across the corpus
▹Semantic search over voice data, not keyword matching

Built with

PythonSpeech-to-textEmbeddingsVector search

Source in a private repo