Call transcription

Call transcription turns the audio of a call into written text. The transcript can be produced live during the call or from the recording afterwards. A live transcript gives agents and supervisors in-the-moment support; a post-call transcript is used for search, summaries, quality review and reporting. Its quality depends on how well the speech-to-text model handles the language, accents and the quality of the phone line.

How it works

The audio is split into segments and fed to a speech recognition model, which returns text and timing for each segment. If the parties were recorded on separate channels, each line is attributed to its speaker directly; with a single-channel recording, that job falls to speaker diarization. Numbers, dates and names are then cleaned up, and personal data can be masked where required.

Why it matters

Searching a transcript takes minutes where listening to an hour of audio takes an hour. The transcript is the basis for the call summary, topic analysis and CRM notes. It is not error-free, though: accuracy is measured with word error rate, and critical details such as amounts, dates or ID numbers should be checked against the recording when they matter. A transcript contains personal data too, so it should follow the same access and retention rules as the recording.