text-to-speech

name: text-to-speech description: Convert text to speech using the slng.ai Voice AI API. Use when generating audio from text, creating voiceovers, synthesizing speech, building voice apps, or working with Deepgram Aura, Rime Arcana, Cartesia Sonic, Sarvam Bulbul, Kugel, Murf Falcon, or Soniox TTS via slng. license: MIT compatibility: Requires internet access and a slng.ai API key (VOICEAI_API_KEY).

slng Text-to-Speech

Generate speech from text using slng's unified Voice AI API. One endpoint reaches every provider (Deepgram, Rime, Cartesia, Sarvam, Kugel, Murf, Soniox) with HTTP one-shot or WebSocket streaming.

Setup: setup-api-key configures VOICEAI_API_KEY. Installation details for each path are in references/installation.md.

Quick Start

CLI

The fastest way — pick a voice, give it text, get an MP3:

voiceai tts "Your support ticket has been resolved." \
  --voice aura-2-thalia-en \
  --out ticket-resolved.mp3

Python

import os, requests

r = requests.post(
    "https://api.slng.ai/v1/tts/slng/deepgram/aura:2",
    headers={"Authorization": f"Bearer {os.environ['VOICEAI_API_KEY']}"},
    json={
        "text": "The package you ordered is out for delivery today.",
        "model": "aura-2-luna-en",
    },
)
r.raise_for_status()
open("delivery-update.wav", "wb").write(r.content)

TypeScript

const r = await fetch("https://api.slng.ai/v1/tts/slng/deepgram/aura:2", {
  method: "POST",
  headers: {
    Authorization: `Bearer ${process.env.VOICEAI_API_KEY}`,
    "Content-Type": "application/json",
  },
  body: JSON.stringify({
    text: "Welcome back. You have three new messages.",
    model: "aura-2-orion-en",
  }),
});
await Bun.write("welcome.wav", await r.arrayBuffer());

cURL

curl https://api.slng.ai/v1/tts/slng/deepgram/aura:2 \
  -H "Authorization: Bearer $VOICEAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"text":"Reminder: your appointment is at three pm.","model":"aura-2-asteria-en"}' \
  --output reminder.wav

Models

slng exposes TTS through a unified endpoint shape: POST /v1/tts/{provider}/{model}:{variant}.

List currently-deployed models from the CLI:

voiceai models --tts

See references/voices-and-models.md for the full catalog and voice-selection guidance.

Picking a voice

The model field in the request body is the voice id (not the model id, which is in the URL path). Format for Aura 2: aura-2-{name}-{lang}.

# List all voices for a TTS model
voiceai voices --model slng/deepgram/aura:2-en

Recommended defaults:

• English feminine: aura-2-luna-en (warm), aura-2-thalia-en (clear)
• English masculine: aura-2-orion-en
• Spanish feminine: aura-2-carina-es
• Spanish masculine: aura-2-sirio-es

Streaming

For long text or real-time playback, use the WebSocket interface:

voiceai tts "long passage..." --stream | ffplay -nodisp -autoexit -

See references/streaming.md for the WebSocket protocol and SDK examples.

Audio output

Default response is binary audio. The CLI writes MP3 by default; the REST API returns WAV. Format depends on the upstream model — Deepgram Aura returns WAV/MP3, Cartesia and Soniox stream raw PCM frames over WebSocket.

Regions

Pin a region to reduce latency or meet data residency requirements:

voiceai tts "..." --region eu-north-1

Available regions: ap-south-1, ap-southeast-2, asia-south1, asia-southeast2, australia-southeast1, eu-north-1, us-east-1. Or set a world part: ap, eu, eu-non-eu, me, na.

Error handling

See also

• references/installation.md — install the CLI, Python SDK, TypeScript SDK
• references/streaming.md — WebSocket streaming across all 4 paths
• references/voices-and-models.md — full voice catalog