Why streaming text-to-speech changes voice agents
Most people judge a voice agent in the first second. If the agent pauses for two seconds before replying, callers talk over it, repeat themselves or hang up.
Where the delay comes from
A typical voice agent chains three systems: speech recognition, a language model and text-to-speech. If each one waits for the previous one to finish completely, the delays add up.
What streaming changes
Streaming text-to-speech accepts text in pieces and returns audio in pieces. The first phrase can play while the language model is still writing the rest of the answer.
What to measure
- Time to first audio: from the moment text is sent to the first sound the listener hears.
- Interruption time: how quickly speech stops when the listener starts talking.
- Pronunciation accuracy: numbers, names and currency, measured on your own scripts.
This article is sample content. Replace it with your own insights before launch.