POST /v1/tts

curl -sS --fail-with-body -X POST https://tts.mayaresearch.ai/v1/tts \
  -H "Authorization: Bearer $MAYA_API_KEY" \
  -H "content-type: application/json" \
  -d '{
        "text": "नमस्ते! आपका ऑर्डर कल पहुँच जाएगा।",
        "voice": "Ananya",
        "language": "hi"
      }' \
  --output out.pcm

import requests

session = requests.Session()
r = session.post(
    "https://tts.mayaresearch.ai/v1/tts",
    headers={"Authorization": f"Bearer {API_KEY}"},
    json={
        "text": "नमस्ते! आपका ऑर्डर कल पहुँच जाएगा।",
        "voice": "Ananya",
        "language": "hi",
    },
)
r.raise_for_status()
pcm = r.content

{
  "error": "unknown_voice",
  "voice": "ananaya",
  "available": ["Ananya", "Arjun"]
}

curl -sS --fail-with-body -X POST https://tts.mayaresearch.ai/v1/tts \
  -H "Authorization: Bearer $MAYA_API_KEY" \
  -H "content-type: application/json" \
  -d '{
        "text": "नमस्ते! आपका ऑर्डर कल पहुँच जाएगा।",
        "voice": "Ananya",
        "language": "hi"
      }' \
  --output out.pcm

import requests

session = requests.Session()
r = session.post(
    "https://tts.mayaresearch.ai/v1/tts",
    headers={"Authorization": f"Bearer {API_KEY}"},
    json={
        "text": "नमस्ते! आपका ऑर्डर कल पहुँच जाएगा।",
        "voice": "Ananya",
        "language": "hi",
    },
)
r.raise_for_status()
pcm = r.content

{
  "error": "unknown_voice",
  "voice": "ananaya",
  "available": ["Ananya", "Arjun"]
}

POST https://tts.mayaresearch.ai/v1/tts

One request, one clip. The audio still streams back as it is generated — you just get one request/response, with no turns and no barge-in.

Body

string

required

The whole thing, in one request. No SSML, HTML or markdown — tags are read aloud literally.

string

default:"Ananya"

Ananya or Arjun. Case-sensitive — the model interpolates whatever it is given, so ananya produces different audio.

string

One of the eleven language codes. Auto-detected when omitted — leave it out for code-mixed text.

string

default:"Maya 2 Native"

Which model answers. Maya 2 Native (default), Maya 2 Native Emotional (supports emotion tags), or Maya 2 Global. See Models.

Response

audio/L16; rate=24000; channels=1

Raw PCM — 16-bit little-endian, mono, 24000 Hz. No file header.

The body is binary. Decoding it as UTF-8 corrupts it, and writing it straight to a .wav file produces something no player will open — see Quickstart for the one-line conversion.

Emotion tags

Set "model": "Maya 2 Native Emotional" and write a tag inline:

{
  "text": "[excited] मुझे यह सुनकर बहुत अच्छा लगा।",
  "voice": "Ananya",
  "language": "hi",
  "model": "Maya 2 Native Emotional"
}

Ten tags — [whispers] [excited] [frustrated] [nervous] [sad] [curious] [laughs] [sighs] [clears throat] [coughs]. Full list and caveats in Emotion tags.

Reuse the connection

The single biggest thing you control:

	First audio
New connection per request	389 ms
Connection reused	88 ms

curl -sS --fail-with-body -X POST https://tts.mayaresearch.ai/v1/tts \
  -H "Authorization: Bearer $MAYA_API_KEY" \
  -H "content-type: application/json" \
  -d '{
        "text": "नमस्ते! आपका ऑर्डर कल पहुँच जाएगा।",
        "voice": "Ananya",
        "language": "hi"
      }' \
  --output out.pcm

import requests

session = requests.Session()
r = session.post(
    "https://tts.mayaresearch.ai/v1/tts",
    headers={"Authorization": f"Bearer {API_KEY}"},
    json={
        "text": "नमस्ते! आपका ऑर्डर कल पहुँच जाएगा।",
        "voice": "Ananya",
        "language": "hi",
    },
)
r.raise_for_status()
pcm = r.content

{
  "error": "unknown_voice",
  "voice": "ananaya",
  "available": ["Ananya", "Arjun"]
}

​Body

​Response

​Emotion tags

​Reuse the connection

Body

Response

Emotion tags

Reuse the connection