Marimba

Multilingual typed audio decisions. Upload or record audio, select questions, and inspect each decision and its probabilities.

Only the last eight seconds are scored; Speech reads the last two seconds using the package's validated default threshold. Speaker change and people count require eight valid seconds; shorter clips return no prediction for those tasks. People counts speakers, not everyone physically present. Language probabilities are relative to the listed choices; Other groups the remaining supported languages. Term probabilities are uncalibrated fractions of five hypotheses (greedy + four beams). No real-time guarantee.

Examples are synthetic speech generated with ElevenLabs. Do not submit sensitive audio; uploads are temporarily cached, with no retention or deletion guarantee.