Faster responses
Responses now average about 160 ms, and 95% return in about 200 ms. With live audio, each answer arrives closer to the moment it describes, and you can send clips more often. These times don’t include the network between your servers and ours.
Shorter audio
A request now takes audio from 1 second long, down from 5 seconds. Audio under 1 second is rejected with audio_too_short, and POST /v0/decisions/file/input_tokens accepts an audio_duration_ms of 1,000 or more. Most questions still give better answers with 5 seconds or more: each use case now shows the shortest audio it holds up on. See Limits.
Higher rate limit
Each organization can now send 600 requests per minute, up from 60. The limit still counts requests to every Audio Decisions endpoint over a sliding 60-second window and is shared by every API key in the organization. See Limits.
Audio Decisions API launch
The Audio Decisions API is now available. Send an audio file with questions written in plain language, and get back probabilities that your code can threshold, rank or combine. Start with the overview.
- Decisions endpoint.
POST /v0/decisions/fileanswers your questions about one audio file in one request. Each question has one of four primitives,noul,choice,scoreorspan, which sets what its answer describes. - Input tokens endpoint.
POST /v0/decisions/file/input_tokensestimates the input tokens of a decisions request. See Billing. - Run endpoints. Runs return the status, timing, audio size and request of each past decisions request.
- SDKs. Python (
hume-audio-decisions) and TypeScript (@humeai/audio-decisions) SDKs cover every endpoint. See the Python quickstart and the TypeScript quickstart.