Speech-to-text

Speech-to-text, also called automatic speech recognition, listens to audio and produces text. In a voice agent that text goes to a language model (often with natural language understanding, which extracts the caller's intent and details such as dates and names), and the reply is turned back into speech by text-to-speech (TTS). The quality of each stage decides how natural the call feels.

For a growing business the stakes are accuracy on your vocabulary: product names, street names in your service area, accents and noisy environments such as a shop floor or a car. Poor recognition produces wrong bookings and frustrated callers; good recognition also gives you searchable transcripts and meeting notes your team actually uses.

Ask a vendor to test recognition on your own terms and callers, whether custom vocabulary can be added, how it handles Spanish or other languages your customers speak, what the delay is between the caller finishing and the agent replying, and which voices are available for text-to-speech. Hear the result on a real phone line, not a laptop speaker.