Screen calls in real time using streaming speech recognition, natural language understanding, and AI voice synthesis. Runs entirely on the device across mobile, embedded, and desktop.
On-device AI call screening automatically answers incoming calls, transcribes what the caller says in real time, understands their intent, and presents the phone owner with action options — with no audio ever sent to a cloud service. Picovoice's Cheetah Streaming Speech-to-Text, Rhino Speech-to-Intent, and Orca Text-to-Speech compose into a self-contained pipeline. All three run locally on the device.
Cheetah Streaming Speech-to-Text beats Google Cloud STT in word error rate and word emission latency across all tested languages, and outperforms Azure STT in several benchmarks, per open-source real-time transcription benchmark — even before it's customized for the use case. Cheetah requires less compute than any other local engine tested.
Most voice command systems run a two-step pipeline: speech-to-text converts audio to a transcript, then a separate NLU model parses that transcript for intent. Every step accumulates error and compounds latency. Rhino Speech-to-Intent is an end-to-end engine with a single model that maps spoken audio directly to a structured intent with typed slot values. Higher accuracy even in noisy environments. No hallucinations. No intermediate transcript.
Most high-quality TTS solutions require hundreds of megabytes of RAM. Orca TTS uses 29 MB peak memory, 10–50× less than any other on-device alternative, except for ESpeak. This makes Orca the only natural-sounding TTS deployable in any environment, including browser tabs, mobile apps with strict out-of-memory limits, and embedded devices.
Mobile OEMs and device manufacturers can ship the same or even more advanced capabilities compared to Google Pixel Call Screening and Apple's native call screening. Picovoice SDKs run on the end-user device. No backend to operate, no privacy agreements to sign. Just a competitive edge over Big Tech.
On-device AI call screening goes beyond STIR/SHAKEN, which only validates the caller ID, and understands what the caller actually says, allowing users to classify their intent in real time. Deploy it as a differentiated feature in your dialer app, or as an upgrade to legacy IVR trees that still cost you DTMF licensing fees.
Healthcare, legal, and FSI teams can improve employee productivity without extra security steps. Keeping caller audio on-device eliminates an entire category of compliance obligation: no BAA for voice data, no processing agreement with a cloud provider, no breach surface through Picovoice systems.
Smart intercoms and access control panels can screen visitor voice queries directly on embedded devices such as Raspberry Pi. No latency from network hops, no dependency on external uptime, no breach risks that contradict the premise of a physical security application. Full call screening capability with no cloud footprint.
A complete working recipe in Python. Open-source on GitHub. Runs 100% on-device.