VoiceAI SDK for Python

The official Python SDK for SLNG Voice AI. Use one client for text-to-speech, speech-to-text, and model discovery across SLNG-hosted and provider-backed speech models.

The package supports Python 3.9 and newer, with sync and async clients powered by httpx.

Install

Until the Python package is published to PyPI, install it from GitHub:

pip install git+https://github.com/slng-ai/voiceai-sdk-python.git

For local development in this repository:

cd sdks/slng-python
uv sync --all-extras

API Key

Create an API key in the SLNG dashboard, then set it in your environment:

export SLNG_API_KEY="zpka_..."

from voiceai_sdk import Slng

client = Slng()

You can also pass api_key directly:

client = Slng(api_key="zpka_...")

Text To Speech

Generate audio from text and save it locally:

from voiceai_sdk import Slng

client = Slng()

audio = client.text_to_speech.create(
    model_variant="slng/deepgram/aura:2-en",
    text="Hello from SLNG.",
    voice="aura-2-thalia-en",
)

audio.write_to_file("hello.wav")

Use region or world_part when you need explicit routing:

client.text_to_speech.create(
    model_variant="slng/deepgram/aura:2-en",
    text="Hello from Europe.",
    voice="aura-2-thalia-en",
    region="eu-north-1",
)

Speech To Text

Transcribe a local audio file:

from pathlib import Path
from voiceai_sdk import Slng

client = Slng()

transcript = client.speech_to_text.create(
    model_variant="slng/deepgram/nova:3-en",
    audio=Path("meeting.wav"),
)

print(transcript.text)

File uploads accept bytes, path-like objects, or (filename, contents, media_type) tuples.

Discover Models

The SDK ships with a static catalog snapshot, so agents and scripts can choose valid models without credentials or network calls.

from voiceai_sdk import get_model, list_models

tts_models = list_models(service="tts", language="en")
stt_models = list_models(service="stt")
aura = get_model("slng/deepgram/aura:2-en")

print([model["id"] for model in tts_models])
print(aura["deployments"] if aura else None)

Model IDs from the catalog can be passed directly to client.text_to_speech.create() or client.speech_to_text.create().

Discover Voices

Use list_voices() to find voice IDs for a TTS model:

from voiceai_sdk import list_voices

voices = list_voices("slng/deepgram/aura:2-en", language="en")

print([f"{voice['name']}: {voice['voiceId']}" for voice in voices])

list_voices() returns an empty list when a model does not have a cataloged voice list.

Useful voice references:

• Voices in the SLNG docs
• Deepgram Aura voices
• Cartesia Sonic 3 voices
• Kugel voices
• Murf Falcon voices
• Orpheus voices
• Rime Arcana voices
• Sarvam Bulbul voices
• Soniox TTS voices
• Full voice catalogs and samples are available in the SLNG dashboard.

Async Usage

Use AsyncSlng for async applications:

import asyncio
from voiceai_sdk import AsyncSlng


async def main() -> None:
    client = AsyncSlng()
    audio = await client.text_to_speech.create(
        model_variant="slng/deepgram/aura:2-en",
        text="Hello from async SLNG.",
        voice="aura-2-thalia-en",
    )
    await audio.write_to_file("hello.wav")


asyncio.run(main())

For lower-level response streaming, use .with_streaming_response on SDK resources.

Reference

• Full API surface: api.md
• General docs: docs.slng.ai
• Models by language: docs.slng.ai/models/by-language
• Source: github.com/slng-ai/voiceai-sdk-python