Semantic Kernel - Lokutor

Semantic Kernel’s OpenAI audio connectors accept a pre-built client, so a base URL override is enough.

Python

from openai import AsyncOpenAI
from semantic_kernel.connectors.ai.open_ai import (
    OpenAIAudioToText, OpenAIAudioToTextExecutionSettings,
    OpenAITextToAudio, OpenAITextToAudioExecutionSettings)
from semantic_kernel.contents import AudioContent

client = AsyncOpenAI(api_key=os.environ["LOKUTOR_API_KEY"], base_url="https://api.lokutor.com/v1")
tts = OpenAITextToAudio(ai_model_id="tts-1", async_client=client)
stt = OpenAIAudioToText(ai_model_id="whisper-1", async_client=client)

audio = await tts.get_audio_contents(
    "Hello from Lokutor.", OpenAITextToAudioExecutionSettings(voice="alloy", response_format="wav"))
open("out.wav", "wb").write(audio[0].data)
text = await stt.get_text_contents(
    AudioContent.from_audio_file("out.wav"), OpenAIAudioToTextExecutionSettings(language="en"))

Notes: the Python voice field only accepts OpenAI voice names, which Lokutor maps onto its voices (alloy and the others map onto F1-F5 / M1-M5). Audio is always WAV and STT takes WAV only. Verified against the live API (TTS to STT round trip) with the semantic-kernel Python package. Runnable copy: lokutor-ai/openai-compatible-examples.

.NET

OpenAITextToAudioService and OpenAIAudioToTextService take an OpenAIClient; build it with new OpenAIClientOptions { Endpoint = new Uri("https://api.lokutor.com/v1") } and your Lokutor key. Not run: the .NET connectors were not tested.