Pronunciations
Teach a character's voice how to say names and words it gets wrong, in dictionaries you attach to characters.
A pronunciation tells the voice how to say a word: "Nguyen" as "Win", "SQL" as "sequel". Entries live in named dictionaries you attach to characters, so one list of your product's names serves every character.
The dashboard's Pronunciations page is coming soon. Today, use the API, or ask your AI assistant through our MCP server (set_pronunciations).
Make a dictionary
Create a pronunciation dictionary with a name and its entries:
curl https://talk.madein10.com/v1/pronunciations \
-H "Authorization: Bearer $SECRET_KEY" \
-H "Content-Type: application/json" \
-d '{ "name": "Names", "entries": [{ "text": "Nguyen", "say": "Win" }, { "text": "SQL", "say": "sequel", "whole_word": false }] }'{
"id": "prn_2Wd7Kq4ZmR9xHc5PvB1n",
"name": "Names",
"entries": [
{ "text": "Nguyen", "say": "Win", "whole_word": true },
{ "text": "SQL", "say": "sequel", "whole_word": false }
],
"character_ids": ["chr_4fT9kQ2mXw7LbN3pZs8V"],
"created_unix": 1793300000,
"updated_unix": 1793300000
}Prop
Type
No text twice in a dictionary, in any case (400 invalid_field).
Spelled as said
say is an alias: write the word the way it sounds. Phonetic alphabets (IPA) aren't offered.
Attach dictionaries to a character
Set a character's pronunciation dictionaries with the whole ordered list; [] detaches them all.
{ "ids": ["prn_2Wd7Kq4ZmR9xHc5PvB1n"] }On a call
- The attached dictionaries' entries apply in order; for a word that appears twice, the first entry wins. At most 200 entries reach a call.
- They change only what the voice says. The transcript, and what the character reads, keep the word as written.
- Where two entries match, the longer match wins.
- A tier that speaks with a speech-to-speech model's own voice ignores them.
Limits
| Dictionaries per organization | 50 |
| Dictionaries per character | 5 |
| Entries per dictionary | 200 |
| Entries on a call | 200 |