Skip to content

Call transcription with AI summaries for contact center calls

Get searchable transcripts and an AI call summary for $0.001/min, then read key details back to callers in a clear text to speech voice.

~200 ms first partial transcript4.76% WER on phone audio$0.001/min
Voice the notice that goes with your transcriptsSpeak example
Lee called Northside Support about order 48213. Dana arranged a replacement. Follow up on Tuesday.
98 characters

Preview an AI call summary

Call transcription turns phone conversations into text, either live while the call runs or from a recording afterwards. PyAI Hear streams a first partial transcript in about 200 ms and handles call recording transcription at 92 to 247 times real time for $0.001/min, with speaker labels and timestamps you can search. Recap then writes an AI call summary with the reason, outcome and next step, so agents stop typing wrap up notes. Because PyAI works as voice infrastructure, one account covers recognition, summaries, the voice and the phone line. PyAI does not replace your ticketing system or CRM, so summaries post to your systems through signed webhooks.

How call transcription and summaries flow

Audio goes in, text and a summary come out, and both land where your team works. The transcript, summary and webhook are committed together, so a ticket never gets a summary without its transcript.

1Call audio

Live stream or recording sent to Hear.

2Transcript

Text with speakers and timestamps.

3Summary

Recap writes the wrap up.

Routine call

PostedSummary sent to the ticket

StoredTranscript saved for search

Needs follow up

TaskNext step created for the agent

FlagTrace marks private data said aloud

4Searchable history

Every call as text, linked to the customer.

Call transcription use cases

Transcripts and summaries cut wrap up time and make every past call easy to find. Support desks, sales teams, clinics and legal intake lines use the same pipeline, with batch jobs running 92 to 247 times faster than real time on long recordings.

Wrap up

Skip manual call notes

The summary is ready when the call ends.

Recap writes reason, outcome and next step, and the agent checks it in seconds.

Search

Search every call

Find a phrase across thousands of calls.

Transcripts with timestamps let teams find a complaint, a promise or a product name quickly.

CRM

Notes in the ticket

Summaries posted by webhook.

A signed HTTPS webhook sends each summary to the ticket or CRM record.

Live

Live captions for agents

See the words as the caller speaks.

Live transcription helps agents on noisy lines and supports caller notes during the call.

Archive

Batch transcription of old recordings

Turn a backlog into text.

Upload archived recordings and process them in bulk at a fraction of the call length. Compliance teams often start here to make a year of stored calls searchable before an audit.

Speakers

Split agent and caller

Know who said each line.

Speaker diarization or a channel split labels the agent and the customer separately, so reviewers can read only the customer side when looking for a complaint or a commitment.

Call transcription cost calculator

Set the recordings a month, the audio length and how long a recording takes your staff by hand. The total covers transcription.

Estimated monthly cost$80.00a month for 20,000 recordings, 80,000 min

Transcription (Hear + Trace)$80.00

667 hagent wrap up time saved each month

Speech to text accuracy on phone audio

Phone calls are narrowband and noisy, so transcription has to be tested on that audio. PyAI Hear posts a 4.76 percent word error rate on 8 kHz phone, accented and noisy audio, starts live text in about 200 ms and runs batch 92 to 247 times real time. Recap is available on any plan.

~200 msfirst partial transcript
92 to 247xreal time batch transcription
$0.001/minHear speech to text
Any planTrace and Recap available
Recording consentFederal law needs one party's consent to record, while a group of states, including Illinois, Maryland and Washington, need every party to agree. Say the call is recorded and transcribed. This is not legal advice.
Redact what you do not needTrace can flag card numbers and other PII so teams remove them from stored transcripts.
BAA on the Enterprise planTeams that handle health information can sign a Business Associate Agreement on the Enterprise plan before patient data touches the API.
Encrypted call recordsRecordings, transcripts and knowledge files sit in encrypted storage scoped to your organization, so no other workspace can read your calls.

Recording consent and privacy for call transcripts

Transcripts hold everything a caller said. Handle them with care.

Build a call transcription from your own code

Build it yourself on Hear + Trace, with three ways in from your own code. Full docs: PyAI SDKs.

SDK

Point the OpenAI SDK at PyAI by swapping the base URL, and Hear + Trace runs your call transcription with no rewrite.

Your keys and limits stay the same across Python and JavaScript on the team.

pip install pyai

MCP

Over MCP, an assistant gains Hear + Trace as a tool and runs your call transcription inside the conversation.

Reviewing a call becomes one tool call among the others the assistant already makes.

pyai mcp add

CLI

Render the spoken notice that goes with your call transcription with the CLI, one command per line.

The CLI shares your SDK keys, so a terminal test moves into code with no new setup.

pyai speak "..."

Call transcription FAQ

What is call transcription?

Call transcription converts spoken phone conversations into written text. It can run live during the call or on a recording after it ends. Contact centers use transcripts for notes, search, quality checks and compliance, and many pair them with an AI summary that lists the reason, outcome and next step.

How does an AI call summary work?

After the transcript is ready, an AI model reads it and writes a short summary. PyAI Recap lists why the customer called, what was agreed and what happens next. The summary can be sent to a ticket or CRM so the next person reads it before calling back.

How accurate is call recording transcription on phone audio?

PyAI Hear is tested on 8 kHz phone, accented and noisy audio, with a 4.76 percent word error rate on that set and 1.59 percent on clean audio. Names and numbers should still be checked for critical details such as account changes.

How much does call transcription cost?

PyAI Hear costs $0.001/min. For example, 12,500 calls of four minutes each make 50,000 minutes, about $50 to transcribe. Recap summaries are on any plan, with no seat license or monthly fee. New accounts get $50 credit with no credit card, and sandbox keys never bill.

Can call transcripts separate the agent from the customer?

Yes. PyAI Hear transcription jobs support speaker diarization on mono recordings and a channel split on stereo recordings. Each line of the transcript is labeled with its speaker and a timestamp, so a supervisor can jump to what the customer said without reading the agent's side.

How do call summaries reach my helpdesk or CRM?

Summaries arrive through a signed webhook when the transcription job completes. Your endpoint receives the transcript, the Recap summary and any Trace flags in one delivery, then posts them to the ticket or contact record. Failed deliveries retry with backoff, so a short outage on your side does not lose a summary.

Start building with PyAI voice

Spin up a sandbox key and put a natural voice on your calls and apps in minutes. No credit card, and $50 in free credit.

Get a sandbox key →