AirCaps AI
API for speech-to-text, diarization and speaker ID that stays accurate in noisy real-world audio.
About AirCaps AI
AirCaps AI: Speech Recognition for Real-World Audio
AirCaps AI is a speech API that transcribes audio with speaker labels, detects who spoke when, and recognizes enrolled voices. Its models are built for real-world audio and stay accurate through background noise, distance and overlapping speakers. All models sit behind one REST API, billed per audio hour, with 5 free hours on every account.
Key Features
Speaker-Attributed Transcription: A5S Async v1 returns transcripts with speaker labels, word timestamps, subtitles and webhooks, in English, Spanish, French, German, Portuguese, Hindi, Japanese, Korean and Chinese.
Speaker Diarization: Diarization-v1 detects speaker turns across meetings, calls and broadcasts without transcribing, and works in any language.
Speaker Identification: Speaker ID-v1 enrolls voices and recognizes them later, even through noise and music, so transcripts can carry the names of enrolled speakers.
Built for Difficult Audio: The models are trained and evaluated for background noise, far-field microphones and crosstalk rather than clean studio speech.
Fast Batch Processing: Models and inference engine are built together, and an hour of audio comes back in about 10 seconds according to AirCaps.
Streaming Options: A5S Streaming v1 delivers live transcripts with speaker labels over WebSocket (early access, on request), and A5S Small V2 is a compact streaming model for noisy, far-field audio.
Flexible Deployment: Cloud API by default, with private cloud (VPC), on-premise and on-device deployment for volume customers; A5S Small V2 also runs air-gapped.
Open Benchmarks: Code, data recipes, raw outputs and scores are published for every result, so accuracy claims can be reproduced.
Data Controls: Accounts can opt out of model training and choose retention settings, and enterprise customers can request zero data retention.
Use Cases
Meetings and Calls: Producing who-said-what transcripts of meetings, sales calls and interviews recorded in noisy rooms.
Media and Broadcast: Generating speaker-labelled subtitles and transcripts for podcasts, video and broadcast content.
Known-Speaker Recognition: Recognizing returning users or team members by voice inside an application.
Private Environments: Running speech recognition in a VPC, on-premise or air-gapped setup where audio cannot leave the network.
Pricing
A5S Async: $0.18 per audio hour for transcripts with speaker labels
Diarization: $0.08 per audio hour
Speaker ID: $0.02 per audio hour
Free Tier: 5 free hours on every account, shared across all models; prepaid credits start at $5 and usage is billed to the millisecond
Volume: Custom pricing from 10,000 hours a month, with higher limits, deployment options and an evaluation on your own audio
Getting Started
Website: https://aircaps.ai
Documentation: https://aircaps.ai/docs
Console: https://console.aircaps.ai
AirCaps AI gives developers transcription, diarization and speaker identification through a single API, tuned for the noisy, multi-speaker audio that real products have to handle.
