Pronunciation dictionaries
Update a pronunciation dictionary
Update a dictionary's display name or atomically replace all of its pronunciation entries.
/pronunciations/v1/workspaces/{workspaceId}/pronunciationDictionaries/{pronunciationDictionaryId}Updating pronunciations replaces the complete entry list. Entries omitted from the request are deleted, and an empty list clears the dictionary.
The request must include the current etag. If another writer has changed the dictionary, the stale etag returns HTTP 409 Conflict instead of overwriting the newer state. Get the dictionary again, reconcile the changes, and retry with the new value.
The primary example supplies both mutable fields and sets updateMask=displayName,pronunciations. The pronunciations value must contain the complete desired dictionary; entries omitted from it are deleted.
Update selected fields
Always send an explicit updateMask query parameter in REST requests. Its comma-separated field names use the same lowerCamelCase spelling as the JSON request body. For example, rename a dictionary without changing its entries:
curl --request PATCH \
"https://api.inworld.ai/pronunciations/v1/workspaces/${INWORLD_WORKSPACE_ID}/pronunciationDictionaries/${PRONUNCIATION_DICTIONARY_ID}?updateMask=displayName" \
--header "Authorization: Basic ${INWORLD_API_KEY}" \
--header "Content-Type: application/json" \
--data "{\"displayName\":\"Product names v2\",\"etag\":\"${PRONUNCIATION_DICTIONARY_ETAG}\"}"Supported fields are displayName and pronunciations. Use updateMask=displayName,pronunciations or updateMask=* only when you intend to update both fields and supply both values. Keep etag in the request body; it is a concurrency token, not an update-mask field.
ID of the workspace that owns the dictionary. It must match a workspace accessible to the API key.
Non-blank human-readable dictionary name, containing at most 64 Unicode code points.
pronunciationsobject[]default: []
Complete dictionary contents. A dictionary can contain up to 1,000 entries and must remain within the 512 KiB encoded dictionary limit. Entries must be unique after language-aware lexical canonicalization.
Show child attributes
Caller-facing spelling. It must be non-blank and tokenize to exactly one replaceable lexical token.
Concrete language supported for pronunciation entries, in its exact canonical BCP-47 spelling: for example, en-US, not en-us, EN_US, or en. Blank values and auto are not accepted. The general TTS language list is not a pronunciation catalog. See Author pronunciation entries.
Ordered array of complete phone tokens supported for the selected language, not a slash-delimited or whitespace-separated IPA string. Each token can contain at most 16 Unicode code points. Preserve compound-phone token boundaries; do not split by Unicode character. Supported aliases may be canonicalized in the response. See phone-token guidance and examples.
Service-assigned resource name in the form workspaces/{workspace}/pronunciationDictionaries/{uuid}.
Service-assigned dictionary ID, the last segment of name. Select a dictionary for synthesis by this ID or by the full name.
Non-blank human-readable dictionary name, containing at most 64 Unicode code points.
pronunciationsobject[]requireddefault: []
Complete dictionary contents. A dictionary can contain up to 1,000 entries and must remain within the 512 KiB encoded dictionary limit. Entries must be unique after language-aware lexical canonicalization.
Show child attributes
Caller-facing spelling. It must be non-blank and tokenize to exactly one replaceable lexical token.
Concrete language supported for pronunciation entries, in its exact canonical BCP-47 spelling: for example, en-US, not en-us, EN_US, or en. Blank values and auto are not accepted. The general TTS language list is not a pronunciation catalog. See Author pronunciation entries.
Ordered array of complete phone tokens supported for the selected language, not a slash-delimited or whitespace-separated IPA string. Each token can contain at most 16 Unicode code points. Preserve compound-phone token boundaries; do not split by Unicode character. Supported aliases may be canonicalized in the response. See phone-token guidance and examples.
Optimistic-concurrency token. Include the current value when updating or deleting the dictionary.