Docs/REST API

REST · text-to-speech.

A single endpoint that takes a sentence and returns audio. Use it for batch jobs, non-realtime workloads, and any system that already speaks HTTP.

Endpoint

POSThttps://api.leanvoice.ai/v1/audio/speech

Returns the audio as the response body. Default response is a WAV file; pass format to switch.

Request body

inputstringRequired

Text to synthesise. Up to 4 000 characters per request. Expression tags are accepted inline.

voicestringRequired

One of the ten voice ids from the voice library. Defaults to meera if omitted.

languagestringOptional

ISO 639-1 language code. Auto-detected from the input if omitted.

formatstringOptional

One of wav, mp3, opus, pcm_s16le_24k, pcm_s16le_16k, mulaw_8k. Default wav.

stepsintegerOptional

Synthesis quality. 4 (draft, fastest), 8 (balanced, default), 10 (studio).

speedfloatOptional

Speaking rate. 0.5 to 2.0, default 1.0.

modelstringOptional

Model identifier. Currently aayush-tts-1 (default); pinned for backwards compatibility.

Response

Audio bytes in the requested format. The following headers carry per-request telemetry:

Header	Meaning
X-Audio-Duration	Length of synthesised audio in seconds, to three decimals.
X-Request-Id	Opaque request id, include in support tickets.
X-Characters-Billed	Characters that counted toward your monthly quota.

Examples

curl

curl https://api.leanvoice.ai/v1/audio/speech \
  -H "Authorization: Bearer $LEANVOICE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"voice":"arjun","input":"Welcome to channel seventeen.","format":"mp3"}' \
  --output greeting.mp3

python

from leanvoice import LeanVoice
client = LeanVoice(api_key=os.environ["LEANVOICE_API_KEY"])
audio = client.speech.create(voice="arjun", input="Welcome.", format="mp3")
open("out.mp3","wb").write(audio.bytes)

node

const audio = await client.speech.create({ voice:"arjun", input:"Welcome.", format:"mp3" });
await writeFile("out.mp3", Buffer.from(await audio.arrayBuffer()));

OpenAI compatibility

The endpoint accepts OpenAI's exact audio/speech request shape, so any client built for OpenAI can switch by changing one URL. Use the openai SDK with base_url set to https://api.leanvoice.ai/v1:

python

from openai import OpenAI

client = OpenAI(
    api_key=os.environ["LEANVOICE_API_KEY"],
    base_url="https://api.leanvoice.ai/v1",
)

audio = client.audio.speech.create(
    model="aayush-tts-1",
    voice="meera",
    input="Same SDK, new base URL.",
)
audio.stream_to_file("out.wav")

The mapping is one-to-one: voice names match the OpenAI shape where compatible (alloy aliases to meera, onyx to arjun, etc.) and our extended fields (steps, speed) ride along as additional parameters that OpenAI's SDK passes through unchanged.

← Previous

Expression tags

WebSocket streaming