Pronunciation dictionary

Alania-2 reads text the way it is spelled in Turkish, so English words and brand names are respelled the Turkish way before they are spoken: PatientDesk becomes “Peyşınt Desk”, ChatGPT “Çet Ci Pi Ti”. A reviewed list of common names and an automatic respeller already run on the server. Words that are yours, such as your clinic's name, doctors' names or drug brands, you add to the pronunciation dictionary.

Matching#

• Whole words only: an entry for Desk does not change Desktop.
• Case does not matter: PatientDesk, patientdesk and PATIENTDESK all match the same entry.
• Terms can be several words: Mayo Clinic → Meyo Klinik.
• Turkish suffixes after an apostrophe are kept: PatientDesk'e → Peyşınt Desk'e, PatientDesk'in → Peyşınt Desk'in.
• An entry whose say_as is the same as its term means “read it as written, don't respell it”. Use it for a word the automatic respeller takes for a foreign one.

Precedence#

When a word appears in more than one place, the first one in this order wins:

1. 1The request's pronunciations field.
2. 2The pronunciation dictionary saved to your account.
3. 3The reviewed built-in list on the server.
4. 4The automatic respeller.

The saved dictionary#

The saved dictionary applies on its own to every speech request from the account, whether it comes with an API key or from the Studio. Your requests need nothing extra.

List#

{
  "object": "list",
  "data": [
    { "id": "b2d4e6f8-1a3c-4e5f-8a7b-9c0d1e2f3a4b", "object": "pronunciation", "term": "PatientDesk", "say_as": "Peyşınt Desk", "created_at": 1759654800 },
    { "id": "7e0a3c51-2f64-4b8d-a1c9-5d2e8f4b6a03", "object": "pronunciation", "term": "Ozempic", "say_as": "Ozempik", "created_at": 1759651200 }
  ]
}

created_at is a Unix time in seconds.

Add or change#

A new word returns 201 Created. If the same word is already saved (ignoring case), its spelling is updated and the response is 200 OK; no duplicate is created.

HTTP/1.1 201 Created
content-type: application/json

{ "id": "b2d4e6f8-1a3c-4e5f-8a7b-9c0d1e2f3a4b", "object": "pronunciation", "term": "PatientDesk", "say_as": "Peyşınt Desk", "created_at": 1759654800 }

Delete#

{ "id": "b2d4e6f8-1a3c-4e5f-8a7b-9c0d1e2f3a4b", "object": "pronunciation", "deleted": true }

Examples#

# your saved list
curl https://voice.patientdesk.ai/v1/pronunciations -H "Authorization: Bearer pd_live_..."

# add a word, or change how a saved one is said
curl https://voice.patientdesk.ai/v1/pronunciations \
  -H "Authorization: Bearer pd_live_..." \
  -H "Content-Type: application/json" \
  -d '{ "term": "PatientDesk", "say_as": "Peyşınt Desk" }'

# delete one by its id
curl -X DELETE https://voice.patientdesk.ai/v1/pronunciations/b2d4e6f8-1a3c-4e5f-8a7b-9c0d1e2f3a4b -H "Authorization: Bearer pd_live_..."

In one request#

For words that belong to one request, add pronunciations to the POST /v1/audio/speech body. They apply to that request only and take precedence over the same words in the saved dictionary. Send an object ({"term": "say as"}) or an array of {term, say_as}.

curl https://voice.patientdesk.ai/v1/audio/speech \
  -H "Authorization: Bearer pd_live_..." \
  -H "Content-Type: application/json" \
  --output hosgeldiniz.wav \
  --data-binary @- <<'JSON'
{
  "model": "alania-v2",
  "voice": "alania2-f1",
  "input": "PatientDesk'e hoş geldiniz. Ozempic kullanıyorsanız lütfen belirtin.",
  "pronunciations": { "PatientDesk": "Peyşınt Desk", "Ozempic": "Ozempik" }
}
JSON

"pronunciations": [
  { "term": "PatientDesk", "say_as": "Peyşınt Desk" },
  { "term": "Ozempic", "say_as": "Ozempik" }
]

In the developer panel's API console you can add pronunciations to a request and hear the result straight away.

Limits#

Errors#