[{"data":1,"prerenderedAt":4},["ShallowReactive",2],{"article-doc:docs\u002Fconcepts\u002Findex":3},"---\ntitle: Voice Context Layer\ndescription: A Voice Context Layer sits between live speech recognition and the application that consumes it.\n---\n\n# Voice Context Layer\n\nA Voice Context Layer sits between live speech recognition and the application that consumes it.\n\n**Realtime**, **Segment**, and **Batch jobs** are three HTTP\u002FWS planes of this layer:\n\n| Plane | Who segments speech |\n|---|---|\n| Realtime WS | Server (VAD on PCM stream) |\n| Segment HTTP | **Client** (you POST each utterance clip) |\n| Batch jobs | Server (time-based slicing of `audio_url`) |\n\nIts job is not simply to produce words.\n\nIts job is to turn continuously changing speech into **usable context**.\n\n## Speech recognition is only the first layer\n\nA speech recognition system may produce something like:\n\n```text\nI think we should meet on\nI think we should meet on Thursday\nI think we should meet on Thursday afternoon\nI think we should meet Thursday afternoon instead\n```\n\nEach output may be reasonable at the moment it was generated.\n\nBut an application has different questions:\n\n- Which text should already be visible?\n- Which text may still change?\n- When is a thought complete enough to translate?\n- When should downstream logic act on it?\n- How should corrections appear without disrupting the reader?\n\nThose questions exist above recognition itself.\n\nThat is the problem addressed by the Voice Context Layer.\n\n## From audio to usable context\n\nConceptually, a live speech system can be viewed as:\n\n```text\nAudio\n  ↓\nSpeech recognition\n  ↓\nVoice Context Layer\n  ↓\nApplication\n```\n\nThe recognition layer determines what was likely spoken.\n\nThe Voice Context Layer determines how that evolving information should become usable by the application.\n\nThis may include:\n\n- stabilization\n- contextual correction\n- segmentation\n- translation readiness\n- presentation continuity\n- lifecycle state\n\n## Why this matters\n\nFor offline transcription, the system can wait until the audio has finished before producing the final result.\n\nLive applications cannot.\n\nThey must continuously balance two competing goals:\n\n**Responsiveness**\n\nShow useful information as soon as possible.\n\n**Stability**\n\nAvoid repeatedly changing information the user has already read.\n\nLansonAI is designed around that tradeoff.\n\nThe objective is not simply to make text appear faster.\n\nIt is to make live speech **ready to use while it is still happening**.\n",1790059118949]