Config reference - SyncWords API documentation

Every field of a project’s config on build 0.9.0-alpha1-2026-10-20T14-10Z-3f05aa, generated from that build’s JSON Schema: the same schema GET /schema returns, and the one the API validates against. The guides explain how the pieces fit together; this page lists all of them. Objects with several kinds (engines, voices, outputs) are told apart by @type, and each kind takes only its own fields. A field marked deprecated is still accepted; do not rely on it.

Another build may differ. GET /schema?buildVersion=… answers the schema of any build, and buildVersion of a project names the build it runs.

The config

Field	Type	Default
`sourceMedia`	one of Input folder, RSS feed, Media RSS feed		Where the media comes from: the project’s input folder (`S3FolderSource`, also what a config without one gets), an RSS feed (`RssSource`: the media in each item’s `<enclosure>`) or a Media RSS feed (`MrssSource`: each item’s `<media:content>`). Set it on create: a change between the folder and a feed is refused; a feed’s `url` and its kind may change.
`transcription`	one of Speechmatics, Soniox, Sarvam	required	What is spoken in the media, and the engine that transcribes it. `@type` picks the engine; left out, the first engine that transcribes `language` is used.
`translations`	array of Translation		One entry per target language: its translation engine and, to dub it, `ttsOptions`. Left out or empty, the media is only transcribed.
`outputs`	array of one of Output folder, Output RSS feed, Output Media RSS feed	required	Where the results go: the project’s output folder (`S3FolderOutput`), and for a feed project one re-published feed per language (`RssOutput` with an `RssSource`, `MrssOutput` with an `MrssSource`). Each lists its `items`, and each item is one file written for every input.

`sourceMedia`

In sourceMedia, by @type.

Input folder

"@type": "S3FolderSource" Files put into the project’s input folder (storage.inputPath), through the API or the AWS role; each is processed once. No fields besides @type.

RSS feed

"@type": "RssSource" An RSS 2.0 feed (a podcast feed, say), read every 30 seconds. Each item’s media is its <enclosure>, downloaded into the input folder, named after the item, and processed once; an item whose media URL changes is processed again. Files cannot be uploaded to such a project.

Field	Type	Default
`url`	string	required	The feed: `https://` only, no credentials in it, on a host that resolves to a public address. A new URL is fetched and read when the project is saved.

Media RSS feed

"@type": "MrssSource" A Media RSS feed, read every 30 seconds. Each item’s media is its <media:content> (the default one of a <media:group>), downloaded into the input folder, named after the item, and processed once; an item whose media URL changes is processed again. Files cannot be uploaded to such a project.

Field	Type	Default
`url`	string	required	The feed: `https://` only, no credentials in it, on a host that resolves to a public address. A new URL is fetched and read when the project is saved.

`transcription`

In transcription, by @type.

Speechmatics

"@type": "Speechmatics" Broad language coverage; the default for most languages.

Field	Type	Default
`language`	string	required	The language spoken in the media, as a code from `GET /enums/languages` (`en`, `es`, `pt-BR`).
`outputLabel`	string		The label written where an output shows one, such as a subtitle track’s name. Default: the language’s English name.
`diarization` deprecated	boolean	`true`	Ignored: speakers are always told apart.
`maxDelay`	number		Seconds the recognizer may take to settle a word; longer is more accurate. Left out, it is chosen for you. From 0 to 4.
`filterDisfluency`	boolean	`true`	Drop “uh”, “um” and similar from the transcript.
`filterProfanity`	string	`"remove"`	`remove` takes profanity out of the transcript; `keep` leaves it in. One of `keep`, `remove`.
`volumeThreshold`	number		Ignore audio quieter than this, so background noise is not transcribed. From 0 to 100.
`punctuationSensitivity`	number		How readily punctuation is added: lower is more conservative. Left out, the language’s own default (usually 0.5). From 0 to 1.
`previewModel`	boolean	`false`	Use Speechmatics’ preview model.
`dictionary`	string or integer		A dictionary of your account in the transcription language, by id or name: words the engine should recognise, such as names and terminology. Managed with the Live API (`/v2/dictionaries`). Left out, your account’s default for the language applies, if it has one; `null` applies none.
`postprocessingDictionary`	one of object, integer, string		Word replacements applied to the transcript before translation, so a name is right in every language. The key is the word you want; the value lists what the engine writes instead, as literals or Java regular expressions. Matching ignores case and applies to single words. An empty key deletes the words it lists. Or a postprocessing dictionary of your account in the transcription language, by id or name, managed with the Live API (`/v2/postprocessing-dictionaries`). Left out, your account’s default for the language applies, if it has one; `null` applies none.
`expectedLanguages`	array of string		With `language: "mul"`, the languages to expect in the media, as codes.

Soniox

"@type": "Soniox" Strong accuracy, and the engine for mixed-language media (language: "mul").

Field	Type	Default
`language`	string	required	The language spoken in the media, as a code from `GET /enums/languages` (`en`, `es`, `pt-BR`).
`model`	string	`"stt-rt-v5"`	The Soniox model.
`extraPrompt`	string		Context that guides recognition: the topic, names, terminology.
`outputLabel`	string		The label written where an output shows one, such as a subtitle track’s name. Default: the language’s English name.
`diarization` deprecated	boolean	`true`	Ignored: speakers are always told apart.
`filterDisfluency`	boolean	`true`	Drop “uh”, “um” and similar from the transcript.
`dictionary`	string or integer		A dictionary of your account in the transcription language, by id or name: words the engine should recognise, such as names and terminology. Managed with the Live API (`/v2/dictionaries`). Left out, your account’s default for the language applies, if it has one; `null` applies none.
`postprocessingDictionary`	one of object, integer, string		Word replacements applied to the transcript before translation, so a name is right in every language. The key is the word you want; the value lists what the engine writes instead, as literals or Java regular expressions. Matching ignores case and applies to single words. An empty key deletes the words it lists. Or a postprocessing dictionary of your account in the transcription language, by id or name, managed with the Live API (`/v2/postprocessing-dictionaries`). Left out, your account’s default for the language applies, if it has one; `null` applies none.
`expectedLanguages`	array of string		With `language: "mul"`, the languages to expect in the media, as codes.

Sarvam

"@type": "Sarvam" Specialised in Indian languages.

Field	Type	Default
`language`	string	required	The language spoken in the media, as a code from `GET /enums/languages` (`en`, `es`, `pt-BR`).
`outputLabel`	string		The label written where an output shows one, such as a subtitle track’s name. Default: the language’s English name.
`diarization` deprecated	boolean	`true`	Ignored: speakers are always told apart.
`filterDisfluency`	boolean	`true`	Drop “uh”, “um” and similar from the transcript.
`dictionary`	string or integer		Not available for Sarvam.
`postprocessingDictionary`	one of object, integer, string		Word replacements applied to the transcript before translation, so a name is right in every language. The key is the word you want; the value lists what the engine writes instead, as literals or Java regular expressions. Matching ignores case and applies to single words. An empty key deletes the words it lists. Inline only: your account’s postprocessing dictionaries apply to Speechmatics and Soniox.
`model`	string	`"saaras:v4"`	The Sarvam model.
`prompt`	string		Terminology hint applied to the final transcript.
`silenceDuration`	seconds	`"0.3"`	The pause, in seconds, that ends an utterance. Shorter pauses give smaller, earlier chunks. Written as a string: `"0.3"`, `"10"`, `"1:30"`.
`expectedLanguages`	array of string		With `language: "mul"`, the languages to expect in the media, as codes.

`translations`

In translations[].

Translation

One target language.

Field	Type	Default
`language`	string	required	The target language, as a code from `GET /enums/languages`.
`engine`	one of NoTranslation, Soniox translation, DeepL, Google Translate, OpenAI, Nova, Gemini		The translation engine. Left out, the recommended engine for the language is used, and stored in the project.
`ttsOptions`	Dubbing		Dub this language. Left out, the translation is used by subtitle outputs only.
`outputLabel`	string		The label written where an output shows one. Default: the language’s English name.

In translations[].engine, by @type.

NoTranslation

"@type": "NoTranslation" No machine translation: the source text is passed through.

Field	Type	Default
`glossary`	one of object, integer		Terms with a fixed translation, `{"source term": "translation"}`: names, brands, terminology. Or the id of a glossary of your account, from the transcription language into this one, managed with the Live API (`/v2/glossaries`).

Soniox translation

"@type": "Soniox" Pairs with Soniox transcription.

Field	Type	Default
`glossary`	one of object, integer		Terms with a fixed translation, `{"source term": "translation"}`: names, brands, terminology. Or the id of a glossary of your account, from the transcription language into this one, managed with the Live API (`/v2/glossaries`).

DeepL

"@type": "DeepL" Fast and literal; excellent for European languages.

Field	Type	Default
`useContext` deprecated	boolean	`true`	Ignored: always on.
`glossary`	one of object, integer		Terms with a fixed translation, `{"source term": "translation"}`: names, brands, terminology. Or the id of a glossary of your account, from the transcription language into this one, managed with the Live API (`/v2/glossaries`).

Google Translate

"@type": "Google" Google Translate: the widest language coverage.

Field	Type	Default
`glossary`	one of object, integer		Terms with a fixed translation, `{"source term": "translation"}`: names, brands, terminology. Or the id of a glossary of your account, from the transcription language into this one, managed with the Live API (`/v2/glossaries`).

OpenAI

"@type": "OpenAI" OpenAI GPT: natural, context-aware translation.

Field	Type	Default
`model`	string	`"GPT-5.6 Luna"`	The model. One of `GPT-5.4 mini`, `GPT-5.6 Luna`, `GPT-5.6 Luna (Thinking)`.
`glossary`	one of object, integer		Terms with a fixed translation, `{"source term": "translation"}`: names, brands, terminology. Or the id of a glossary of your account, from the transcription language into this one, managed with the Live API (`/v2/glossaries`).
`extraPrompt`	string	`"Avoid profane language in translated output."`	Instructions for the translator: style, register, audience. They replace the default, so restate it to keep it.
`latencyMode`	string		`quality` translates whole sentences; `low-latency` smaller units. Left out, it is chosen to suit the transcription engine. One of `quality`, `low-latency`, `null`.
`compression`	boolean	`true`	Condense long translations, which keeps a dub in step with the original speech.
`enableSpeakerGenderAware` deprecated	boolean	`true`	Deprecated by the build; kept so older configs stay valid.

Nova

"@type": "Nova" Amazon Nova: natural, context-aware translation.

Field	Type	Default
`model`	string	`"Nova 2 Lite"`	The model. One of `Nova 2 Lite`.
`glossary`	one of object, integer		Terms with a fixed translation, `{"source term": "translation"}`: names, brands, terminology. Or the id of a glossary of your account, from the transcription language into this one, managed with the Live API (`/v2/glossaries`).
`extraPrompt`	string	`"Avoid profane language in translated output."`	Instructions for the translator: style, register, audience. They replace the default, so restate it to keep it.
`latencyMode`	string		`quality` translates whole sentences; `low-latency` smaller units. Left out, it is chosen to suit the transcription engine. One of `quality`, `low-latency`, `null`.
`compression`	boolean	`true`	Condense long translations, which keeps a dub in step with the original speech.
`enableSpeakerGenderAware` deprecated	boolean	`true`	Deprecated by the build; kept so older configs stay valid.

Gemini

"@type": "Gemini" Google Gemini: natural, context-aware translation. A strong default for dubbing.

Field	Type	Default
`model`	string	`"Gemini 3.1 Flash Lite"`	The model. One of `Gemini 3.1 Flash Lite`, `Gemini 3.1 Flash Lite (Thinking)`.
`glossary`	one of object, integer		Terms with a fixed translation, `{"source term": "translation"}`: names, brands, terminology. Or the id of a glossary of your account, from the transcription language into this one, managed with the Live API (`/v2/glossaries`).
`extraPrompt`	string	`"Avoid profane language in translated output."`	Instructions for the translator: style, register, audience. They replace the default, so restate it to keep it.
`latencyMode`	string		`quality` translates whole sentences; `low-latency` smaller units. Left out, it is chosen to suit the transcription engine. One of `quality`, `low-latency`, `null`.
`compression`	boolean	`true`	Condense long translations, which keeps a dub in step with the original speech.
`enableSpeakerGenderAware` deprecated	boolean	`true`	Deprecated by the build; kept so older configs stay valid.

In translations[].ttsOptions.

Dubbing

How a translation is dubbed: the voice, and how the new voice is mixed with the original.

Field	Type	Default
`provider`	one of Amazon Polly, Azure, Deepdub, Google Text-to-Speech, MultiVoice, Sarvam voice cloning	required	The voice provider and the voice.
`locale`	string		The pronunciation locale, such as `es-XL` (Latin American Spanish) or `pt-BR`. It must belong to the translation’s language. Left out, it is chosen from the language.
`volume`	number	`1`	Level of the new voice. From 0 to 2.
`originalAudioSplitting`	boolean	`true`	`true` separates the original into voices and background, keeps the background and removes the original voices: a clean dub. `false` keeps the whole original under the new voice, at `originalAudioVolume`: a voice-over.
`originalAudioVolume`	number	`1`	Level of what is kept of the original: the background, or the whole original for a voice-over. `0` leaves it out; `1` is its own level. From 0 to 10.
`lateUtteranceTimeout`	seconds	`"10"`	Seconds after which a line that could not be synthesised in time is skipped. Written as a string: `"0.3"`, `"10"`, `"1:30"`.
`dictionary`	object		How words are spoken: `{"SyncWords": "sink words"}` has “SyncWords” said as “sink words”.

In translations[].ttsOptions.provider, by @type.

Amazon Polly

"@type": "Amazon" Amazon Polly voices, by name. GET /voices?provider=Amazon lists known ones.

Field	Type	Default
`voice`	string	required	The voice, by its id or name in the provider’s catalogue. It must speak the target language. Only a Deepdub voice is checked when the project is saved; for Amazon, Azure and Google, `GET /voices` lists known ones. See Choosing a voice.
`engine`	string	`"NEURAL"`	The Polly engine; not every voice has every engine. One of `STANDARD`, `NEURAL`, `GENERATIVE`.
`rate`	integer	`100`	Speaking rate in percent, used with `rateStrategy: "static"`. From 20 to 200.
`rateStrategy`	string	`"dynamic"`	`dynamic` speeds each line up or slows it down to fit the time the original took; `static` keeps `rate`. One of `static`, `dynamic`.
`minDynamicRate`	integer	`80`	The slowest `dynamic` may go, in percent. From 50 to 200.
`maxDynamicRate`	integer	`160`	The fastest `dynamic` may go, in percent. From 50 to 200.

Azure

"@type": "Azure" Microsoft Azure neural voices, by name. GET /voices?provider=Azure lists known ones.

Field	Type	Default
`voice`	string	required	The voice, by its id or name in the provider’s catalogue. It must speak the target language. Only a Deepdub voice is checked when the project is saved; for Amazon, Azure and Google, `GET /voices` lists known ones. See Choosing a voice.
`rate`	integer	`100`	Speaking rate in percent, used with `rateStrategy: "static"`. From 20 to 200.
`rateStrategy`	string	`"dynamic"`	`dynamic` speeds each line up or slows it down to fit the time the original took; `static` keeps `rate`. One of `static`, `dynamic`.
`minDynamicRate`	integer	`80`	The slowest `dynamic` may go, in percent. From 50 to 200.
`maxDynamicRate`	integer	`160`	The fastest `dynamic` may go, in percent. From 50 to 200.

Deepdub

"@type": "Deepdub" Clones each speaker’s own voice into the target language.

Field	Type	Default
`voice`	string	required	Always `auto`, which clones each speaker’s own voice; anything else is refused.
`model`	string	`"syncwords34-baked"`	The Deepdub model.
`backupModel`	string	`"dd-etts-3.4"`	The model used when `model` is unavailable.
`promptBoost`	boolean	`false`	Work harder to reproduce each speaker’s characteristics; less robust.
`cleanMode`	boolean	`false`	More robust with noisy source audio; somewhat the opposite of `promptBoost`. Not every model has it.
`referenceBufferDuration`	seconds	`"2"`	Seconds of each speaker’s audio used as the reference for cloning. Written as a string: `"0.3"`, `"10"`, `"1:30"`.
`accentLocale`	string		Blend in an accent: a Deepdub locale, mixed in by `accentRatio`.
`accentRatio`	number	`0.75`	How much of `accentLocale` to blend in: 0 is fully native, 1 fully the accent. From 0 to 1.
`rate`	integer	`100`	Speaking rate in percent, used with `rateStrategy: "static"`. From 20 to 200.
`rateStrategy`	string	`"dynamic"`	`dynamic` speeds each line up or slows it down to fit the time the original took; `static` keeps `rate`. One of `static`, `dynamic`.
`minDynamicRate`	integer	`80`	The slowest `dynamic` may go, in percent. From 50 to 200.
`maxDynamicRate`	integer	`160`	The fastest `dynamic` may go, in percent. From 50 to 200.

Google Text-to-Speech

"@type": "Google" Google Cloud Text-to-Speech voices, by name. GET /voices?provider=Google lists known ones.

Field	Type	Default
`voice`	string	required	The voice, by its id or name in the provider’s catalogue. It must speak the target language. Only a Deepdub voice is checked when the project is saved; for Amazon, Azure and Google, `GET /voices` lists known ones. See Choosing a voice.
`rate`	integer	`100`	Speaking rate in percent, used with `rateStrategy: "static"`. From 20 to 200.
`rateStrategy`	string	`"dynamic"`	`dynamic` speeds each line up or slows it down to fit the time the original took; `static` keeps `rate`. One of `static`, `dynamic`.
`minDynamicRate`	integer	`80`	The slowest `dynamic` may go, in percent. From 50 to 200.
`maxDynamicRate`	integer	`160`	The fastest `dynamic` may go, in percent. From 50 to 200.

MultiVoice

"@type": "MultiVoice" A voice for each speaker, chosen for you from SyncWords’ voices for the translation’s language.

Field	Type	Default
`rate`	integer	`100`	Speaking rate in percent, used with `rateStrategy: "static"`. From 20 to 200.
`rateStrategy`	string	`"dynamic"`	`dynamic` speeds each line up or slows it down to fit the time the original took; `static` keeps `rate`. One of `static`, `dynamic`.
`minDynamicRate`	integer	`80`	The slowest `dynamic` may go, in percent. From 50 to 200.
`maxDynamicRate`	integer	`160`	The fastest `dynamic` may go, in percent. From 50 to 200.

Sarvam voice cloning

"@type": "Sarvam" Voice cloning for Indian languages.

Field	Type	Default
`rate`	integer	`100`	Speaking rate in percent, used with `rateStrategy: "static"`. From 20 to 200.
`rateStrategy`	string	`"dynamic"`	`dynamic` speeds each line up or slows it down to fit the time the original took; `static` keeps `rate`. One of `static`, `dynamic`.
`minDynamicRate`	integer	`80`	The slowest `dynamic` may go, in percent. From 50 to 200.
`maxDynamicRate`	integer	`160`	The fastest `dynamic` may go, in percent. From 50 to 200.

`outputs`

In outputs[], by @type.

Output folder

"@type": "S3FolderOutput" The project’s output folder (storage.outputPath). One per project; its items’ names must differ.

Field	Type	Default
`items`	array of one of Video file, Audio file, Subtitle file	required	The files written to the output folder for every input.

Output RSS feed

"@type": "RssOutput" The source RSS feed re-published in one language, at GET /feeds/{id}/{language}/feed.xml: every item whose media is done links the dubbed file in its <enclosure> instead, <channel><language> is set, and the rest of the feed passes through. Items not done yet are left out. Needs an RssSource; one per language.

Field	Type	Default
`language`	string	required	The feed’s language: the transcription language or a translation language. Written to `<channel><language>`.
`items`	array of one of Video file, Audio file, Subtitle file	required	Exactly one video or audio file, which each feed item links to; other items are written beside it (under `output/feeds/{language}/`) and not linked yet.

Output Media RSS feed

"@type": "MrssOutput" The source Media RSS feed re-published in one language, at GET /feeds/{id}/{language}/feed.xml: every item whose media is done links the dubbed file in its <media:content> instead, <channel><language> is set, and the rest of the feed passes through. Items not done yet are left out. Needs an MrssSource; one per language.

Field	Type	Default
`language`	string	required	The feed’s language: the transcription language or a translation language. Written to `<channel><language>`.
`items`	array of one of Video file, Audio file, Subtitle file	required	Exactly one video or audio file, which each feed item links to; other items are written beside it (under `output/feeds/{language}/`) and not linked yet.

In outputs[].items[], by @type.

Video file

"@type": "VideoFile" The input’s video with the audio and subtitle tracks you choose.

Field	Type	Default
`containerFormat`	string	`"mp4"`	The container, as an ffmpeg muxer name. It may differ from `ext`: `matroska` writes `.mkv`.
`ext`	string	required	The file extension: `${ext}` in `nameTemplate`.
`nameTemplate`	string	required	The file’s name. `${name}` is the input’s name without its extension, subfolders included; `${ext}` is `ext`. It must contain `${name}`, and no two items of an output may render the same name.
`useOriginalTracks`	boolean	`false`	`true` copies the input’s audio into the file as it is, with its own language, title and default flag, ahead of the dub tracks.
`dubTracks`	array of Dub track		The dubbed audio tracks, in order, after the input’s audio. With neither, the video has no sound.
`subTracks`	array of Subtitle track		Subtitle streams in the file, in order, which players can select.
`videoCodec`	string	`"copy"`	`copy` keeps the input’s video untouched, the fastest. An ffmpeg encoder (`libx264`) re-encodes it.

Audio file

"@type": "AudioFile" An audio file: MP3, M4A, WAV, …

Field	Type	Default
`containerFormat`	string	required	The container, as an ffmpeg muxer name: `mp3`, `ipod` (for `.m4a`), `wav`, …
`ext`	string	required	The file extension: `${ext}` in `nameTemplate`.
`nameTemplate`	string	required	The file’s name. `${name}` is the input’s name without its extension, subfolders included; `${ext}` is `ext`. It must contain `${name}`, and no two items of an output may render the same name.
`useOriginalTracks`	boolean	`false`	`true` makes the file the input’s audio, copied as it is: AAC in most inputs, so `ipod` (`.m4a`) or `matroska`, not MP3 or WAV. Either this or one dub track.
`dubTracks`	array of Dub track		The dub the file holds: exactly one entry, or none with `useOriginalTracks`.

Subtitle file

"@type": "SubtitleFile" A subtitle or transcript file in one language.

Field	Type	Default
`format`	string	required	The caption format: `SRT`, `VTT`, `TXT`, … Also the file’s extension, in lower case.
`language`	string	required	The transcription language or a translation language.
`nameTemplate`	string		`${name}`, `${lang}` (the file’s language) and `${ext}`. Default: `${name}.${lang}.${ext}`.
`options`	any		Caption layout options, passed to the caption renderer as given.

In outputs[].items[].dubTracks[].

Dub track

One dubbed audio track of a video or audio file.

Field	Type	Default
`language`	string	required	The code of a dubbed translation (one with `ttsOptions`), written as in `translations`.
`codec`	string	`"copy"`	`copy` keeps the dub’s AAC. An ffmpeg encoder re-encodes it: `libmp3lame` for MP3.
`title`	string		The track name players show in their audio menu.

In outputs[].items[].subTracks[].

Subtitle track

A subtitle stream inside the video, which players can select.

Field	Type	Default
`language`	string	required	The transcription language or a translation language.
`options`	any		Caption layout options, passed to the caption renderer as given.
`subtitleCodec`	string	`"mov_text"`	`mov_text` for MP4 and MOV, `subrip` or `ass` for Matroska, `webvtt` for WebM.

​The config

​sourceMedia

​Input folder

​RSS feed

​Media RSS feed

​transcription

​Speechmatics

​Soniox

​Sarvam

​translations

​Translation

​NoTranslation

​Soniox translation

​DeepL

​Google Translate

​OpenAI

​Nova

​Gemini

​Dubbing

​Amazon Polly

​Azure

​Deepdub

​Google Text-to-Speech

​MultiVoice

​Sarvam voice cloning

​outputs

​Output folder

​Output RSS feed

​Output Media RSS feed

​Video file

​Audio file

​Subtitle file

​Dub track

​Subtitle track

The config

`sourceMedia`

Input folder

RSS feed

Media RSS feed

`transcription`

Speechmatics

Soniox

Sarvam

`translations`

Translation

NoTranslation

Soniox translation

DeepL

Google Translate

OpenAI

Nova

Gemini

Dubbing

Amazon Polly

Azure

Deepdub

Google Text-to-Speech

MultiVoice

Sarvam voice cloning

`outputs`

Output folder

Output RSS feed

Output Media RSS feed

Video file

Audio file

Subtitle file

Dub track

Subtitle track