Voice API (TTS)

The Dhamma.Gift voice is an open text-to-speech service: the same one that reads in the site's player and the dictionary. Anyone can call it from a browser or a server — no key, no login.

Address: https://api.dhamma.gift · CORS is open · a limit of 60 requests per minute per IP.

Read a phrase

POST /api/tts/pali with JSON:

Field	Meaning
text	the text, up to 2000 characters. Pāḷi in IAST (Evaṁ me sutaṁ), a translation in English or Russian
voice	the voice (below); pratham by default
rate	speed, 0.25–3; 1 is the voice's own pace (the site reads Pāḷi at 0.875)

The answer is {"audioContent": "<base64 MP3>"} — the same shape as Google's Text-to-Speech, so a player written for Google plays it as is. MP3, mono, 22 kHz, ~40 kbit/s; repeated requests come from a cache.

curl -s https://api.dhamma.gift/api/tts/pali -H 'content-type: application/json' \
  -d '{"text":"Evaṁ me sutaṁ.","voice":"pratham","rate":0.875}' \
  | python3 -c 'import json,sys,base64; open("evam.mp3","wb").write(base64.b64decode(json.load(sys.stdin)["audioContent"]))'

const r = await fetch('https://api.dhamma.gift/api/tts/pali', {
  method: 'POST', headers: { 'content-type': 'application/json' },
  body: JSON.stringify({ text: 'Evaṁ me sutaṁ.', voice: 'pratham', rate: 0.875 })
});
new Audio('data:audio/mp3;base64,' + (await r.json()).audioContent).play();

Voices

voice	Language	What it is
pratham	Pāḷi	Piper neural voice + our Pāḷi pronunciation rules (IAST → sounds, stress, long vowels). Beta
dg	Pāḷi	"o Dhamma.Gift": a voice trained on real Pāḷi readings. Beta
alan, norman, kathleen	English	Piper voices; Pāḷi words and abbreviations (Ven., SN 56.11) are read properly
ruslan, irina	Russian	Piper voices; Pāḷi in Latin script is read the Pāḷi way

GET /api/tts/dg-voices gives the list as {voice: language}; GET /api/tts/health shows that the service is up and which version of the pronunciation rules it uses.

How Pāḷi is read: every letter is sounded, punctuation becomes pauses, …pe… is read as "peyyāla".

A whole text as one MP3

POST /api/tts/memo — the lines read one after another with pauses of any length between them (Google stops at 10 seconds); this is what the Memo page saves.

Field	Meaning
segments	the lines, up to 200 and 20 000 characters in all
voice, rate	as above
delay	the pause between lines, seconds, up to 300
end_delay	the pause at the end, seconds, up to 600
sound	gong.mp3 or tick.mp3 at the end, or empty

The answer is the MP3 itself (audio/mpeg), at most 45 minutes long. One such request counts as ten against the per-minute limit.

Licences

• The service code and the Pāḷi rules: github.com/dhammagift/pali-tts.
• pratham: weights MIT (rhasspy/piper-voices), recordings AI4Bharat IndicVoices (CC BY 4.0).
• dg: CC BY-NC-SA 4.0 — non-commercial use only.
• The English and Russian voices: rhasspy/piper-voices, each under its own licence.

Also at this address

The same host serves the MCP server (/mcp) and the site's API (/api/…), with the same texts and database as dhamma.gift.