Text to speech overview

Chariot converts text into natural speech through three transports that share the same voices, models, and per-character pricing. They differ in latency and how you feed text in.

Transport	Endpoint	Output	Best for
REST (sync)	`POST /v1/tts`	Complete `audio/wav` file	Short text where you want one finished file
HTTP streaming	`POST /v1/tts/stream`	Raw PCM stream	Low time-to-first-byte for a single utterance
WebSocket	`wss://api.chariot.in/v1/tts/ws`	Audio frames per flush	Real-time, incremental, or long-running speech

Not sure which to use? See Choosing an API.

The request

All three take the same core inputs:

string

required

UUID of the voice to synthesize with. Get one from GET /v1/voices.

string

required

The text to speak. After trimming whitespace, sync accepts 1 to 700 characters for chariot-v0 or 1 to 1,000 for chariot-multilingual-v1. HTTP streaming accepts 1 to 500 or 1 to 1,000 respectively. The WebSocket accepts incremental fragments within the same limits as HTTP streaming, including the accumulated unflushed buffer.

string

default:"chariot-v0"

Use chariot-v0 or chariot-multilingual-v1. Pass this in the JSON body for sync and HTTP streaming, or the connection URL query for WebSocket. See Models.

number

default:"1.0"

Select a speaking rate from the voice’s entry for your model in models[].speeds. Chariot v0 supports only 1.0. Pass this as a query parameter on WebSocket connections.

integer

default:"0"

An integer from 0 to 4. Higher values trade quality passes for lower latency. Pass this as a query parameter on WebSocket connections.

Example

This example explicitly selects Chariot Multilingual v1. Choose a voice whose models array includes chariot-multilingual-v1.

curl -X POST https://api.chariot.in/v1/tts \
  -H "chariotai-api-key: YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "voice_id": "bac7d666-094d-4698-91fa-741d60fce662",
    "text": "Chariot turns text into speech.",
    "model_type": "chariot-multilingual-v1"
  }' \
  --output speech.wav

Response metadata

The synchronous endpoint returns audio in the body and metadata in headers:

Header	Description
`Inference-Id`	Unique id for the generation
`Audio-Url`	Persistent URL to the audio
`Credit-Utilized`	Credits charged (= character count)
`Content-Disposition`	Suggested download filename

The streaming endpoint uses X-inference-id and X-audio-url for the same purpose. Its Credit-Utilized header reports the request cost; billing occurs after successful delivery.

Formats and sample rate

Sync (POST /v1/tts): WAV, 44.1 kHz, mono.
Streaming (POST /v1/tts/stream): raw signed 16-bit little-endian PCM (audio/L16), 44.1 kHz, mono, no WAV header. The X-audio-url header points to a WAV archive uploaded after generation. Archiving is best effort, so retain the streamed audio if you need a guaranteed local copy.

Cost

TTS costs 1 credit per input character. A 200-character request costs 200 credits. If a generation fails after you were charged, the credits are refunded. See Credits & pricing.

Next steps

Choosing an API

REST vs. streaming vs. WebSocket, with a decision table.

WebSocket streaming

The full real-time protocol.

Voices

Browse the voice catalog.

Rate limits & concurrency

How many requests you can run at once.

Get started

Text to speech

Speech to text

Resources

Text to speech overview

The request

Example

Response metadata

Formats and sample rate

Cost

Next steps

Choosing an API

WebSocket streaming

Voices

Rate limits & concurrency

​The request

​Example

​Response metadata

​Formats and sample rate

​Cost

​Next steps

Choosing an API

WebSocket streaming

Voices

Rate limits & concurrency

The request

Example

Response metadata

Formats and sample rate

Cost

Next steps