Audio & Voice Pipeline

How dialogue lines become audio -- TTS, voice changer, recording, and merged character tracks.

Ordinary Animator has a full dialogue audio pipeline: from dialogue lines written in the screenplay, through voice synthesis or recording, to per-character merged audio tracks that play back on the sequencer. The episode's whole spoken track lives on the Dialogue Audio page (Dialogue in the episode's menu), where you can generate it in one press and work through it line by line. This page covers the one-press surface; for the small operations underneath it -- and the recipes the platform builds from them -- see Building Dialogue from Operations.

Dialogue lines

Dialogue is written in the screenplay (Fountain format) and parsed into the scene's dialogue structure. Each line belongs to a character, and once you bring it onto the Sequencer it also has a place in time -- a clip you can drag. The episode Dialogue view lists every line in order; the Sequencer is where each one sits.

Which shot a line belongs to is not something you set. A shot is a span between two cuts, and its dialogue is whatever falls inside that span, so dragging a line past a cut moves it to the next shot and moving a cut changes what the shots either side of it say. A shot's Properties view shows what currently falls in it, read-only, with a link to the Sequencer.

Every line can independently have:

  • Generated TTS audio
  • A recorded audio file (uploaded or captured)
  • A voice-changer version -- a recording re-voiced as the character

Voice configuration

Each character has a Voice configuration. You select which TTS engine to use and which voice within that engine:

Engine Notes
Google Gemini Performs the line's parenthetical as delivery direction; a fresh take each run
Google Chirp Fast, consistent, and cheap -- the recommended default for a first pass (shown as "Google" in some pickers)
ElevenLabs High-quality synthesis with many voice options; requires API key
Typecast Alternative TTS engine
ComfyUI Run text-to-speech or a voice changer through your own ComfyUI workflows

A character can have more than one voice set up at once (for example Google Chirp for a quick first cut, plus a voice changer for hero lines), and you mark one as the draft voice from its menu. Set this up on the character's Voice page. The one-press generation uses each character's draft voice, so there's nothing to pick at run time.

Generating dialogue audio

You rarely generate audio one line at a time. The fastest path is one press for everything:

  • On the Dialogue Audio page, and in the Dialogue rendered / recorded step of the episode plan, Generate missing dialogue audio works through every line in the episode and creates the audio for each one that has none. Each scene also has its own Generate missing audio for this scene button, and the same button appears on a scene's page. It only creates a take for each line; it does not build the per-shot timing (see "One shot at a time" below).
  • Once every line has audio, the same button changes: Re-render N changed lines when the script has changed since a line was voiced, otherwise Another take for every line.
  • The button just runs -- it uses each character's starred draft voice, so there's no per-run prompt. Characters whose draft voice is a voice changer, or who have no voice set up, are listed for you rather than synthesised.
  • It only fills in text-to-speech. Lines you intend to record, run through a voice changer, or let a video model voice are left untouched and listed afterwards so you know what still needs a human step -- they are never treated as errors.
  • Re-running is safe: lines that are already up to date are left alone; only new or changed lines are regenerated.

This is built for an audio-first preview: generate the whole episode's voice track in one click, listen end to end, then upgrade individual lines (record a performance, clone a voice) where it matters.

For per-line control -- alternates, recording, voice changing, picking the preferred take -- expand a line on the Dialogue Audio page (its Takes control, which shows the take count once it has audio). You can preview each clip inline, regenerate, and choose between takes.

One shot at a time

When you are working on a single shot rather than the whole episode, its media gallery carries a Render dialogue audio button. It does the same job for that one shot: one audio file per speaking character, timed to the Sequencer, dropped into the shot's gallery and labelled with whose voice it is. Each file runs exactly as long as the shot, with silence wherever that character is not speaking. That per-character file is what a lip-sync video model takes alongside the shot's start image, which is how a shot ends up speaking instead of silent.

Once it has been pressed the button reads Re-render dialogue audio. It turns orange when what is in the gallery no longer matches the Sequencer -- the shot got longer or shorter, a line moved, a take changed, or a character stopped speaking in it. It never re-renders on its own; the colour is the prompt. See Shots.

Reviewing and choosing takes

A line can end up with several takes -- a quick first pass, a higher-quality alternate, a recorded performance, a voice-changer version. You never have to choose: generating a line automatically marks the newest take as the one that's used, so there's always a usable voice. Choosing is optional polish.

Picking is done right where the lines are: on the Dialogue Audio page or the Dialogue rendered / recorded step, each line shows how many takes it has. Expand a line to play each take and choose the one you want -- or generate, record, or clone another take on the spot. Picking a take is the only choice you make per line.

Because a shot's combined per-character audio is built from the chosen takes, changing a choice puts the affected shots' combined audio out of date. The shot's Re-render dialogue audio button turns orange, and the episode preview lists the shot as Needs re-merge; pressing the button rebuilds the shot's audio to match your picks.

Previewing the episode

The Review the episode on the sequencer criterion (in the Sound on the Sequencer step of the episode plan) holds Preview the episode -- the quickest way to hear how everything fits together before you invest in final renders.

  • Prepare preview audio generates any missing dialogue in one press (the same cache-aware run as Generate missing dialogue audio).
  • A per-shot checklist shows each shot as Audio ready, No dialogue, Needs audio or Needs re-merge.
  • Preview episode on the Sequencer opens the Sequencer, where pressing Play lays the dialogue over whatever media each shot has -- a storyboard still or a video -- with any music cues over the top. You don't need a finished video to hear the episode: a storyboard with voices is enough to judge pacing and performance.

Each line is heard once. A talking clip -- a rough clip made from a first frame and a re-voice -- carries the line in its own sound, and the dialogue track carries it again. OA knows whether a video has its own sound (it notes this when the file arrives, and works it out the first time an older video is played on the Sequencer), and a shot plays only one of the two:

  • a clip with its own sound plays that sound, and the shot's dialogue is silent;
  • a clip without sound gets the dialogue track played over it, as before;
  • a video whose own sound has not been checked yet behaves as a silent clip until it has.

Mute any clip with its speaker. Every video clip that has its own sound, and every dialogue, music and effects clip, carries a small speaker on its bar. One tap mutes that clip; tap again to hear it. A muted clip shows a struck speaker and a dimmed waveform. The speakers start where the rule above puts them: on a talking clip the clip's own speaker is on and that shot's dialogue clips show muted; on a silent clip it is the other way round. A video clip with no sound, or whose sound is silent throughout (generated clips often carry a silent track), shows no speaker at all; the first time you open the Sequencer it listens to the clip once to find out, and until then it shows none. A video clip that does have sound draws its waveform on the bar, following the part of the clip the shot plays, so you can see where the voice is. A tap overrides that for the one clip, and the Sequencer, the Shot List reel and the export all follow it, so the finished episode has the same sound you heard in the preview. Because a shot says each line once, turning a dialogue line on under a talking clip turns the clip's sound off for that shot, and turning the clip's sound back on takes back the lines you had switched on. The track's own mute and solo still work as before, on top of this.

A clip longer than its shot shows the footage to spare. Shot lengths are yours to set, and trimming the start and end of a rough clip is normal: the shot plays a window of the clip, and a short clip holds its last frame. Select a video clip on the Video row and a faint shadow shows the whole file, running past the shot's cuts, with a label such as "+1.2 s" or "+2.7 s" for how much footage lies beyond each edge. Nothing is shown when there is none to spare, or for a still image, a looping clip or a file whose length is unknown. Drag inside the selected clip to slide the window along the file: a different part of the clip plays, while the shot, its cuts and its length stay where they are. The drag stops where the footage ends, and the labels update as you drag. The export plays the same part of the clip, picture and sound together. To change how long the shot is, move its cut, as always.

Shots you haven't given artwork yet don't show a black screen: the preview stands in a starred image -- the speaking character's, or failing that the scene's location -- so you can still follow the episode end to end. Where more than one character speaks over the shot, it stands in ALL of them: the frame is split into equal columns, one per speaker, in the order they speak, so a two-hander reads as a two-hander. An export of that shot draws the same split, so what you previewed is what you get.

A shot's own length is yours to set, and dialogue no longer stretches it: a line that runs past a cut simply plays on over the next shot.

Recording and uploads

If you prefer a human voice or want to capture reference audio:

  • Record -- capture audio directly in the browser
  • Upload -- attach an existing audio file

Uploaded and recorded files go through the same curating flow as TTS: you preview, approve, or replace.

Voice changer

A voice changer turns a recorded performance into a character's voice, keeping the original timing and delivery. Set one up on the character's Voice page (ElevenLabs speech-to-speech or a ComfyUI workflow) and attach reference voices as the target timbre. This lets you perform a line yourself -- or in any voice -- and have it come out sounding like the character. It's the path to use when TTS isn't enough and you want a real performance in the character's voice.

Merged audio tracks

Once a shot's dialogue lines are generated or recorded, Render dialogue audio on the shot merges the audio per character. The merge produces a single audio track per character, with silence between spoken lines -- and the gaps are the gaps you set on the Sequencer, so the pacing you listened to is the pacing the shot renders against. Move a line and the merge is reported out of date, because it is: the file was built from where the line used to be.

The merged track is what the sequencer renderer uses when assembling the final video -- it's mixed across all characters in the scene.

The render pipeline

See Shots for how rendered video clips and audio tracks are combined in the shot and scene sequencer render.