Definition
Speech-to-text turns spoken audio into text in near real time, the first stage of translated calls.
Speech-to-text, often shortened to STT or ASR, is the pipeline that converts an audio stream into written words. For live calls the model has to deliver partial transcripts every few hundred milliseconds, cope with pauses and overlapping voices, and recover from truncation when the speaker finishes a sentence. ChromePing uses Deepgram's streaming STT, driven by a LiveKit agent that forwards the Opus audio frames; partial transcripts land in the browser as captions while the final transcript is handed to the machine-translation stage.
Related terms
Text-to-speech (TTS)
Text-to-speech is the last stage of a translated call, it reads the translated text aloud in the target language's voice.
Call translation latency
Call-translation latency is the time between the moment a speaker finishes a sentence and the moment the other party hears the translated version, ideally under 1.5 seconds.
Audio codec
An audio codec compresses voice into a bitstream for transmission. Opus is the browser-standard codec used by WebRTC and ChromePing.