Sentioscope

Speech and voice controls, as each vendor documents them

What Synthesia documents about speech

Synthesia (docs.synthesia.io) publishes a voice catalogue in which every entry carries a language name, a language code, a gender and a voice id, and states that lip sync and facial expressions are generated from the spoken content. From 2 source pages, read 2026-09-22; no maker pays to be listed.

A published number against a published catalogueMost entries in this register give a count of languages. This one gives rows: a formal name, a native name, a language code, a gender and an id for each voice, plus a separate language list for cloned voices.A count in a paragraphA catalogue with codesWhat it answersWhether the feature exists at allWhether one named market is coveredWhat can be automatedNothing; it is proseA request, built from an idTracking a changeSomebody has to reread the sentenceTwo exports can be comparedCloned voicesUsually unaddressedA language list kept separatelyThe difference is operational, not editorial
Fig. 1 A count tells a planner the capability exists; rows tell them whether the market they were asked about is inside it.
Controls documented, model by modelFilled where the model documents that control, hollow where nothing is published about it.Controls documented, model by modelWhat the vendor documentsWhat the vendordocumentsAudio sourceAudio source — What the vendor documents: Synthesised from the script, or an uploaded recordingLanguagesLanguages — What the vendor documents: A catalogue, each voice carrying a language name and codeVoice sourceVoice source — What the vendor documents: The catalogue, or a cloned voice with a language list of its ownPer-character bindingPer-character binding — What the vendor documents: Not documentedLip-syncLip-sync — What the vendor documents: Generated from the spoken content, with a framing condition attached
Fig. 2 Filled where the model documents that control, hollow where nothing is published about it.
The five fields this register keeps, for Synthesia. Read from the vendor's voices reference and avatar documentation on 2026-09-22.
FieldWhat the vendor documents
Audio sourceSynthesised from the script, or an uploaded recording
LanguagesA catalogue, each voice carrying a language name and code
Voice sourceThe catalogue, or a cloned voice with a language list of its own
Per-character bindingNot documented
Lip-syncGenerated from the spoken content, with a framing condition attached

Inclusion rule. Fields are the same five for every model here; a field the vendor does not document is recorded as not documented rather than inferred from generated output. Order. Fixed field order, identical on every model page.

1A language list a schedule can be built on

Instead of a number in a paragraph, this vendor publishes rows: the formal language name, the native name, a language code, a gender, a voice name and an id. A planner can compare that list between quarters, pick a market off it, and pass the id straight into a request.

The register records it as a catalogue rather than a count, because the difference is operational. A count tells a planner the capability exists somewhere; a list with codes tells them whether the market they were asked about is inside it.

2Lip sync is tied to the script, and comes with a condition

Lip sync and facial expressions are generated from the spoken content, which places this entry with the models that derive the mouth from the words rather than repairing it afterwards. Then the documentation adds something almost unheard of here: lip sync performs best when the avatar is framed closer in the scene rather than positioned far from the camera.

That sentence is a shot-list instruction wearing the clothes of a help note. It says the weak case is the wide shot, which is precisely the shot a director reaches for when two people are in a scene together.

3A cloned voice has its own reach

Voice cloning carries a language list of its own, running from Afrikaans to Zulu and described as the same list used for standalone cloning. So the question is not only which languages the platform speaks but which of them a particular voice can speak, and the two lists are kept separately.

What is not published is how a voice stays attached to a character across a run of videos. An id can be reused; nothing states that the pairing is held anywhere except in a production's own notes.

4This entry, one column at a time

Each of these stays inside a single field: what Synthesia puts there, what the wording settles, and what it leaves for a take to answer.

5Read against another entry

Each of these puts Synthesia beside one other entry on a column where the two land at opposite grades of answer.

6Where this model sits against the rest

Audio source, across every model hereOne field at a time, every entry side by side, with Synthesia marked. The full wording for every cell is in the rows below. This entry was read 2026-09-22.Audio source, across every model hereAudio sourceD-ID — Audio source: A text script read aloud, or an audio url suppliedD-IDHeyGen — Audio source: A text to speech endpoint, apart from the pictureHeyGenKling AI — Audio source: Generated with the picture, VIDEO 3.0Kling AILTX Studio — Audio source: Generated jointly with the picture, plus audio-to-videoLTX StudioLuma Ray — Audio source: Not documented; the video reference does not mention soundLuma RayMiniMax — Audio source: Native speech, generated with the pictureMiniMaxPixVerse — Audio source: Uploaded audio, or text read by a chosen voice; the two are not mixedPixVerseRunway — Audio source: Not documented where voices are definedRunwaySceneMixer — Audio source: Generated with the picture, in the language set for the projectSceneMixerSora 2 — Audio source: Generated with the picture; video and audio are both listed as outputSora 2sync-3 — Audio source: Supplied, or read from text on the callsync-3Synthesia — Audio source: Synthesised from the script, or an uploaded recordingSynthesiaVeo — Audio source: Generated with the video rather than added afterwardsVeoVidu — Audio source: Q3 generates speech, effects and music natively, on by defaultViduWan 3.0 — Audio source: On by default; false returns a file with no audio trackWan 3.0Wan2.2-S2V — Audio source: Supplied; the card describes the model as audio-drivenWan2.2-S2V
Fig. 3 One field at a time, every entry side by side, with Synthesia marked. The full wording for every cell is in the rows below. This entry was read 2026-09-22.
Every entry in this register on the same field, with Synthesia marked. Each cell carries the reading date of its own entry; this one was read 2026-09-22.
ModelAudio source
D-IDA text script read aloud, or an audio url supplied
HeyGenA text to speech endpoint, apart from the picture
Kling AIGenerated with the picture, VIDEO 3.0
LTX StudioGenerated jointly with the picture, plus audio-to-video
Luma RayNot documented; the video reference does not mention sound
MiniMaxNative speech, generated with the picture
PixVerseUploaded audio, or text read by a chosen voice; the two are not mixed
RunwayNot documented where voices are defined
SceneMixerGenerated with the picture, in the language set for the project
Sora 2Generated with the picture; video and audio are both listed as output
sync-3Supplied, or read from text on the call
SynthesiaSynthesised from the script, or an uploaded recording
VeoGenerated with the video rather than added afterwards
ViduQ3 generates speech, effects and music natively, on by default
Wan 3.0On by default; false returns a file with no audio track
Wan2.2-S2VSupplied; the card describes the model as audio-driven

Inclusion rule. Models whose vendor documents something about generated speech. A model with nothing published on the point is still listed, with the cell marked hollow. Order. Alphabetical by model name.

Languages, across every model hereOne field at a time, every entry side by side, with Synthesia marked. The full wording for every cell is in the rows below. This entry was read 2026-09-22.Languages, across every model hereLanguagesD-ID — Languages: A language field beside the voice; no list on the referenceD-IDHeyGen — Languages: Thirty or more, counted on the translation endpointHeyGenKling AI — Languages: Five, not namedKling AILTX Studio — Languages: Not documentedLTX StudioLuma Ray — Languages: Not documentedLuma RayMiniMax — Languages: Not documented by the vendorMiniMaxPixVerse — Languages: A languages field on each voice, none listedPixVerseRunway — Languages: Not documentedRunwaySceneMixer — Languages: 15, named, with Cantonese a 16th for dialogue onlySceneMixerSora 2 — Languages: Not documentedSora 2sync-3 — Languages: Ninety-five or more, counted rather than namedsync-3Synthesia — Languages: A catalogue, each voice carrying a language name and codeSynthesiaVeo — Languages: Not documentedVeoVidu — Languages: Not documentedViduWan 3.0 — Languages: Not documentedWan 3.0Wan2.2-S2V — Languages: Not documentedWan2.2-S2V
Fig. 4 One field at a time, every entry side by side, with Synthesia marked. The full wording for every cell is in the rows below. This entry was read 2026-09-22.
Every entry in this register on the same field, with Synthesia marked. Each cell carries the reading date of its own entry; this one was read 2026-09-22.
ModelLanguages
D-IDA language field beside the voice; no list on the reference
HeyGenThirty or more, counted on the translation endpoint
Kling AIFive, not named
LTX StudioNot documented
Luma RayNot documented
MiniMaxNot documented by the vendor
PixVerseA languages field on each voice, none listed
RunwayNot documented
SceneMixer15, named, with Cantonese a 16th for dialogue only
Sora 2Not documented
sync-3Ninety-five or more, counted rather than named
SynthesiaA catalogue, each voice carrying a language name and code
VeoNot documented
ViduNot documented
Wan 3.0Not documented
Wan2.2-S2VNot documented

Inclusion rule. Models whose vendor documents something about generated speech. A model with nothing published on the point is still listed, with the cell marked hollow. Order. Alphabetical by model name.

Voice source, across every model hereOne field at a time, every entry side by side, with Synthesia marked. The full wording for every cell is in the rows below. This entry was read 2026-09-22.Voice source, across every model hereVoice sourceD-ID — Voice source: A voice id from one of five named speech providersD-IDHeyGen — Voice source: A clone, instant from one recording or professional from 20 minutes or moreHeyGenKling AI — Voice source: Not documented by the vendorKling AILTX Studio — Voice source: Not documentedLTX StudioLuma Ray — Voice source: Not documentedLuma RayMiniMax — Voice source: Reference audio, 15 s total across 3 clipsMiniMaxPixVerse — Voice source: Built-in voices, or a custom voice made from uploaded audioPixVerseRunway — Voice source: An audio sample of 10 seconds to 5 minutes, or a description in wordsRunwaySceneMixer — Voice source: Preset voice library, or the user's own recordingsSceneMixerSora 2 — Voice source: Not documentedSora 2sync-3 — Voice source: Not documented on the model pagesync-3Synthesia — Voice source: The catalogue, or a cloned voice with a language list of its ownSynthesiaVeo — Voice source: Not documentedVeoVidu — Voice source: Not documentedViduWan 3.0 — Voice source: Reference audio, 15 seconds in total, WAV or MP3 up to 15 MBWan 3.0Wan2.2-S2V — Voice source: Whatever track is handed over; no catalogueWan2.2-S2V
Fig. 5 One field at a time, every entry side by side, with Synthesia marked. The full wording for every cell is in the rows below. This entry was read 2026-09-22.
Every entry in this register on the same field, with Synthesia marked. Each cell carries the reading date of its own entry; this one was read 2026-09-22.
ModelVoice source
D-IDA voice id from one of five named speech providers
HeyGenA clone, instant from one recording or professional from 20 minutes or more
Kling AINot documented by the vendor
LTX StudioNot documented
Luma RayNot documented
MiniMaxReference audio, 15 s total across 3 clips
PixVerseBuilt-in voices, or a custom voice made from uploaded audio
RunwayAn audio sample of 10 seconds to 5 minutes, or a description in words
SceneMixerPreset voice library, or the user's own recordings
Sora 2Not documented
sync-3Not documented on the model page
SynthesiaThe catalogue, or a cloned voice with a language list of its own
VeoNot documented
ViduNot documented
Wan 3.0Reference audio, 15 seconds in total, WAV or MP3 up to 15 MB
Wan2.2-S2VWhatever track is handed over; no catalogue

Inclusion rule. Models whose vendor documents something about generated speech. A model with nothing published on the point is still listed, with the cell marked hollow. Order. Alphabetical by model name.

7Sources

Taken from the voices reference and avatar documentation at docs.synthesia.io on 2026-09-22. Every entry against the same five fields sits on the speech controls page; the rule behind each field is on how read. The entries either side of this one: Veo, Vidu.

  • Languages
    Each voice is listed with its formal and native language name, a language code, a gender, a name and a voice idpublished as a catalogue with codesSynthesia, list of supported voices / recorded 2026-09-22
  • Lip-sync
    Lip sync and facial expressions are generated from the spoken contentdriven by the scriptSynthesia, create an avatar / recorded 2026-09-22
  • A framing condition attached to lip-sync
    Lip sync performs best when the avatar is framed closer in the scene rather than positioned far from the cameraSynthesia, create an avatar / recorded 2026-09-22
  • Voice source
    Voice cloning carries a language list of its own, running from Afrikaans to Zulu and described as the same list used for standalone cloninga cloned voice with its own reachSynthesia, create an avatar / recorded 2026-09-22