What is Speech-to-text?
Speech-to-text (STT), also called automatic speech recognition (ASR), converts spoken audio into written text.
In more detail
STT is the listening half of a voice agent and the basis of call transcripts, meeting notes and call analytics. Accuracy is reported as word error rate. It drops with background noise, accents, crosstalk, phone-line compression and domain words such as product names or medical terms. Fixes include custom vocabularies, noise suppression and a model trained for telephone audio. Streaming STT returns partial results while people speak, which voice agents need. Batch STT processes a finished recording and is usually more accurate. Stored transcripts can contain personal data, so set retention rules and redact where required.
Example
A sales team records discovery calls. Batch STT produces transcripts overnight, and an LLM summarizes next steps and updates the CRM. A vocabulary list holds the company's product names, which otherwise come out garbled.