Hume Audio Decision API: Analyzing Sound Without Text Transcription

Hume has launched the Audio Decision API, an AI-powered tool for analyzing audio data. The system takes an audio file and a text prompt, returning a structured response along with a confidence score.

Traditional voice processing methods first convert speech to text, which often results in the loss of intonation and pause data. The new API analyzes the raw audio file directly, acting as a decision-making layer for voice agents without requiring intermediate transcription.

The API enables the following tasks:

• Determining the number of speakers in the stream
• Identifying specific sound events, such as laughter
• Analyzing interruptions between interlocutors in a dialogue

For processing long recordings, it is recommended to split files into segments and submit requests in parts. Developers should verify confidence scores on their own datasets, as they do not guarantee absolute accuracy in emotion interpretation.

API access is available upon request via the product page. Documentation and usage examples are available on the company's official resources.