Sentioscope

Speech and voice controls, as each vendor documents them

Where an accent is written when a voice is generated

Three entries document a way to ask for an accent, and each puts the request somewhere different: in the description used to design a stored voice, in the native audio of a video model, or in a line describing a character's voice that the model performs on every line that character speaks. As of 2026-10-13.

Where the accent request sits, and how long it lastsIn a voice-design description the accent is fixed when the voice is created. In a video model's native audio it is asked for on each generation. In a character's voice description it rides on every line the character speaks, until a preset or recording is attached and takes over.Set onceAsked for each timeWhere it is writtenA voice design, or a character's voicelineThe native audio of one generationWhat keeps itThe stored voice, or the unchanged lineThe person writing the next promptWhat overrides itAn attached preset or recordingNothing describedA matched voice replaces the description, accent included
Fig. 1 Two of the three placements are set once; one has to be repeated on every call.
Where each entry lets an accent be requested, and when the request is fixed. Recorded 2026-10-13.
ModelWhere the accent is asked forWhen it is fixed
Kling AIIn the native audio, alongside five languagesPer generation
RunwayIn the description used to design a custom voiceWhen the voice is created
SceneMixerIn the character's voice descriptionOn every line, while no preset or recording is attached

Inclusion rule. Entries whose documentation names accent or dialect as something a request can specify. A language list with no mention of accent does not earn a row. Order. Alphabetical by model name.

1Three places for one request, and three kinds of consistency

Where the accent is asked for decides how it behaves across a season. Written into the description of a voice that is then created and stored, it is fixed once and every later use inherits it. Written into the native audio of a video call, it is asked for again on each generation. Written into a line attached to the character, it travels with every sentence that character speaks without anyone repeating it.

The first and third are set once. The second is a per-call instruction, and a per-call instruction is the kind that gets forgotten in episode nine.

2A recording overrides a description

One entry states the rule plainly: once a preset is matched to the character or a recording is uploaded, the description is no longer what is performed, and an accent written there goes nowhere. Its guide recommends checking for an attached voice before generating.

The same tension is built into the voice-design entry, which offers two separate routes to a voice, a sample or a description of at least twenty characters. An accent belongs to the description route; a sample carries whatever accent its speaker had.

3Accent, dialect and a different language are three settings

One model documents dialects and accents next to five languages and allows dialogue to switch language inside one scene. Another sets the dialogue language per project, from fifteen named languages plus Cantonese, and leaves the accent to the character's own description within that language.

The distinction matters for a production with a mixed cast. A character who speaks the project's language with a regional accent is a description; a character who speaks another language is a language setting, and the two are not interchangeable.

  • Asking for an accent
    Dialects and accents are rendered alongside the five native-audio languages, and dialogue can switch language within one sceneKling AI, model guide / recorded 2026-10-13
  • Audio source
    Native audio on VIDEO 3.0 in five languagesgenerated with the pictureKling AI, model guide / recorded 2026-09-12
  • Asking for an accent
    A voice designed from a description should name tone, accent, pacing and personalityRunway, custom voices reference / recorded 2026-10-13
  • Voice source
    A voice can instead be asked for in words, with the description required to run at least 20 charactersa route that needs no recordingRunway, custom voices reference / recorded 2026-09-22
  • Per-character binding
    A voice is created asynchronously, reaches a ready state with a preview, and is then referred to by its own ida stored voice objectRunway, custom voices reference / recorded 2026-09-22
  • Asking for an accent
    An accent written into a character's voice description is performed on every line that character speaksSceneMixer, accent guide / recorded 2026-10-13
  • What overrides the request
    Once a preset is matched or a recording uploaded, the description is not used, so an accent written there is not performedSceneMixer, accent guide / recorded 2026-10-13
  • A voice written in words
    By default each character's voice is a line of description, covering age, weight, pace and where it sits, which the video model performsSceneMixer, character voices guide / recorded 2026-10-13
  • Languages
    Dialogue is set per project to any of 15 languages, with Cantonese a 16th for dialogue onlya named list of 15SceneMixer, languages guide / recorded 2026-09-12

4Sources

Every line is read from the vendor documentation linked on the speech controls page, checked 2026-10-13. What counts as documented is on how read. Nearby: Changing a line, Second language, Keeping a voice.