🌟 Hume AI Unveils Model That Answers Questions from Voice Recordings
Hume AI has launched the Audio Decision API, the first audio-native model that answers questions about speech and sound directly from raw audio, without transcription. The input consists of a recording, questions, and allowed answers; the output is probabilities with confidence scores: yes/no, multiple choice, a scale, or a time segment. Agreement with the average opinion of a group of listeners is 70.7%, compared to 56.5% for a single evaluator.
🌍 One model replaces dozens of separate systems: diarization, language, emotion, and non-speech events. This reduces the cost of call analysis in contact centers and the development of voice agents; questions can be changed without retraining.
👤 The model understands not only words but also intonation, pauses, laughter, background noise, and the number of speakers. Access is available via a waitlist request, with documentation and SDKs provided.
Source 1: https://www.hume.ai/blog/introducing-hume-audio-decision-api Source 2: https://www.hume.ai/products/decision-model-api