Speech to Text (STT)

Speech to text (STT, speech-to-text) turns audio into text, which is further processed by the model. How to embed: select a mode - streaming (transcription as speech occurs, for live dialogue) or batch (uploaded a file - received text, for recordings); take into account accuracy factors - noise, accent, terms, language (often the language must be specified explicitly); handle recognition errors (STT makes mistakes, especially on names and numbers - let the user check and correct); pass the text on to LLM for meaning. STT is the input, not the end: the raw transcript almost always needs to be cleaned up or made sense of. Rule: STT gives the text, but not the understanding - the model follows the meaning, and monitor the quality by noise, language and error checking.