Features
PyAI voice infrastructure features, from diarization to dialing
PyAI is voice infrastructure: Omni voice agents, Hear speech to text, Speak text to speech, Trace call review, Recap summaries and AMD answering machine detection run on one engine behind one API key. Pricing is per use, starting at $0.001/min for Hear, with no platform fee and a $50 starter credit. Pick a feature below by the job you need done.
Listening and turn taking
How a PyAI voice agent knows when a caller starts, pauses and finishes, including turn detection, silence detection and interruption handling.
Voice activity detection →
Turn detection that ends a caller's turn in about 196 ms, so the agent replies quickly without talking over a slow speaker.
WebRTC voice activity detection →
See how a WebRTC detector distinguishes speech from background sound in a live voice workflow.
Interrupt handling →
Barge-in support: when a caller cuts in, the pending reply is dropped and your app gets a signal to stop playback.
Transcription and speakers
Streaming speech to text for live audio, batch jobs for recordings, and diarization that labels who said each line.
Speaker diarization →
Diarization for meetings, podcasts and phone recordings, returned as labelled speakers from async Hear jobs.
Real time transcription →
Live partial text arrives in roughly 200 ms over a WebSocket, then settles into a final line once the speaker stops.
Call transcription →
Send a finished call file and get text, segments and speaker labels back; an hour of audio finishes in well under a minute.
Voicemail transcription →
Turn voicemails left on your existing phone system into readable text at $0.001/min, with spoken numbers written as digits.
Voice output
Text to speech for apps and voice agents, priced at $0.04/min of audio delivered.
Call control and routing
What happens to a live call after the agent answers: routing rules, handoffs to people, recordings and wait lines.
Call routing →
Each phone number maps to one agent, and that agent's written rules decide which calls ever reach a human.
Call transfer →
Move a live caller from the agent to your team the moment a rule you wrote is met.
Warm transfer →
Your teammate gets context on the caller before joining, so the customer never repeats the story.
Call whisper →
A short private message plays to the person picking up a transfer, before the caller is connected.
Call queue →
Instead of hold music, every caller reaches an agent session straight away and only matched calls ring staff.
Call recording →
Two channel recordings stored encrypted per organization, with a disclosure line spoken before recording starts.
Agents and integrations
Connect Omni agents to the documents, calendars and CRMs your team already uses, and use AI function calling through custom tools.
Knowledge base →
Upload trusted documents once; the agent searches them mid call and answers aloud with a short reply.
Appointment booking →
Callers hear real open slots from your scheduler and leave the call with a confirmed booking.
Calendar integration →
A custom tool behind a signed webhook lets the agent read and write the calendar you already run.
CRM integration →
After the call, a note, ticket or field update lands in HubSpot, Zendesk, Pipedrive, Zoho CRM or a webhook.
Bilingual receptionist →
Run an English agent and a Spanish agent on separate numbers, with handoff to bilingual staff when needed.
Analytics and QA
Measure calls, review them with Trace at $0.12/call in beta, and produce call summarization with Recap at $0.02/min.
Call analytics →
Duration, pickup, cost and failed requests per call, tracked without reading the conversation itself.
Speech analytics →
Hear transcribes the words, then Recap pulls out topics, decisions and promises made during the call.
Sentiment analysis →
PyAI gives no sentiment score; Recap risk signals, trackers and Hear transcripts feed the scorer your team picks.
Emotion detection →
Calls only: PyAI reads words, not vocal tone, so transcripts and Recap flags stand in where an emotion label would go.
Call scoring →
Trace grades every finished call against your scorecard, and each finding links to the exact second.
PII redaction →
Agent replies are screened before they are spoken, and Trace checks finished transcripts for personal data.
Outbound calling
One REST request per call with live keys, plus AMD at $0.004/call to tell people from voicemail. PyAI is not a pacing dialer with rep seats. Not legal advice.
Outbound dialer →
Your code owns the contact list; PyAI handles duplicate protection and checks before each phone rings.
Auto dialer →
A guide to dialer types and TCPA prior express consent, and where single API calls fit instead.
Power dialer →
Keep the one lead at a time sales rhythm, with an agent opening and a human rep closing.
Predictive dialer →
Pacing math and the 3% abandoned call cap explained, next to an approach with an agent on every answer.
PyAI prices and benchmarks checked against PyAI product docs on 29 Oct 2026.
PyAI features FAQ
What is diarization?
Diarization splits a recording by speaker, so each line of the transcript shows who said it. PyAI Hear returns speaker labels from async transcription jobs for meetings, interviews, podcasts and phone calls. The speaker diarization page covers request fields, output format and the $0.001/min price for Hear.
Which PyAI feature handles turn detection?
Voice activity detection handles turn detection inside Omni voice agents. PyAI ends a caller's turn in about 196 ms, so the agent answers fast without cutting in on someone who pauses mid sentence. Interrupt handling covers the opposite case, when the caller talks over the agent.
Do PyAI features cost extra?
No feature carries a separate fee. You pay for the product a feature uses: Omni at $0.05/min plus $0.01 telephony, Hear at $0.001/min, Speak at $0.04/min, Recap at $0.02/min, Trace at $0.12/call in beta and AMD at $0.004/call. There is no platform fee.
How do I start testing PyAI features?
Sign up in the PyAI Console and create a sandbox key. New accounts get $50 starter credit with no credit card required, and sandbox keys never bill. Test calls end after 60 seconds, and placing outbound calls needs a live key with the telephony manage scope.