# Timing & Assembly (https://hypit.ai/quickstart/timing/) > Per-take normalization and alignment, followed by SemanticTrack assembly. For spoken video, a `SemanticTrack` connects the authored Script to the actual performance. This is the natural time source for captions, word-triggered graphics and coverage. Build it in segment-sized pieces: 1. normalize each accepted A/V take into one exact frame domain; 2. align that normalized media with its authored Script Segment to create a self-contained `SemanticTake`; 3. assemble the Semantic Takes in program order with `speech:Track`. Every Take is already semantic before it enters the Speech Track. Pictures can be presented by Media Track or a project component independently of this semantic and audio assembly. A purely visual animation can instead declare its own duration and frame rate in a ProgramSpace. See [an authored film clock](/quickstart/composition#a-film-drawn-entirely-by-components). It can use seconds or frames for events; spoken work can use Script Selections and Moments for the same visual behavior. ```svml ``` ## Normalize each take [#normalize-each-take] Normalization makes video, audio, duration and frame rate one explicit `SynchronizedMedia` fact. The Clock is authored once and shared by every Take that will enter the same SemanticTrack. ```svml ``` Normalization contains no Script meaning and performs no transcription. It only establishes the media facts that later semantic alignment can trust. ## Create one SemanticTake per Segment [#create-one-semantictake-per-segment] `whisperx:SemanticTake` measures one normalized Take and aligns the evidence with exactly one authored Segment: ```svml ``` `language` is required on every alignment call and currently accepts `en` or `zh`. It is passed unchanged to WhisperX; Hypit does not detect or route languages from Script text or audio. Each output contains the normalized media, the Segment identity, every authored word's local frame window, and all of that Segment's structural anchors. There are two anchors for the Segment and two for each word. Acoustic evidence is an implementation input to this step; downstream components see the completed `SemanticTake`, not a second evidence-shaped timing structure. ## Assemble the SemanticTrack [#assemble-the-semantictrack] `speech:Track` concatenates already-semantic Takes in document order and provides the semantic timeline and its original sound. Media presents the performance separately: ```svml ``` | Output | Type | Meaning | | ------------------- | ------------- | ------------------------------------------------ | | `{speech.semantic}` | SemanticTrack | Global semantic and frame-domain authority | | `{speech.audio}` | AudioTrack | Same-source sound, aligned to the semantic items | Picture consumers and the original audio use the same prepared Takes and source positions. `SemanticTrack` derives global frames by prefix-summing the local Take lengths, and also supplies the program duration and frame domain required by Film and Render. ## Consume semantic time [#consume-semantic-time] Selections, Moments and whole Segments remain authored Script identities. A downstream component receives the SemanticTrack once and projects those identities into frames only when it builds its deterministic Track: ```svml ``` Use `during={story.segment.answer}` for a whole Segment, a Selection for an authored range, a Moment for a point event, and `during="program"` for the complete SemanticTrack domain. Components consume `semantic={speech.semantic}`. ```text prepared Takes → Speech Track ── .semantic → Media / project scene → .visual ─┐ │ └──→ Caption / semantic graphics ───┤ └───── .audio ────────────────────────────────────┤ Film ```