Speech in. Speech out.
One API for voice.
Transcribe audio with speaker labels, generate natural speech, and clone a voice — behind one OpenAI- and ElevenLabs-compatible API. Built for voice agents and content creators. Run it in our cloud, on your own servers, or on-device.
Understand speech — and generate it.
Steno turns audio into attributed transcripts. Vox turns text into natural speech, in a preset voice or a clone of your own. Same key, same billing, same endpoint.
Speech-to-text with speaker diarization, speaker identification, and word-level timestamps. Our tuned model, covering 25+ languages and their regional dialects — word-for-word. Auto-detect the language or hint it.
Natural text-to-speech, returned as a complete file or streamed low-latency for live use. Use a preset voice, or clone your own from a short reference clip and reuse it across every call. Drop-in for the OpenAI and ElevenLabs audio APIs.
You address each model by its Soniqo name; we keep improving what runs underneath without you changing a line.
Drop Soniqo into your agent stack.
Steno and Vox speak the OpenAI and ElevenLabs audio APIs — so an agent framework like LiveKit works by pointing it at our base URL. No custom plugin, streaming PCM at the rate agent runtimes expect.
# LiveKit Agents — Soniqo TTS with no custom plugin
from livekit.plugins import openai
tts = openai.TTS(
base_url="https://api.soniqo.audio/v1",
api_key="sk_...", # sent as Authorization: Bearer
model="tts-1",
voice="nova", # a preset, or one of your cloned voices
response_format="pcm", # low-latency 24 kHz stream
)Any stack that already speaks the OpenAI or ElevenLabs audio APIs works the same way — change the base URL, keep your code.
Everything you need to ship voice features.
Diarization, dialect coverage, drop-in compatibility, and a billing model that bundles it all — across transcription, synthesis, and cloning.
Every utterance comes back labelled with a speaker. Register speaker profiles once and get stable named identities across meetings, interviews, and call recordings.
25+ languages with regional-dialect coverage on the core model, plus a long-tail model spanning 1,600+ languages (Hindi, Arabic, Indonesian, Vietnamese, and more). Auto-detect or set an explicit hint.
Code written against the OpenAI or ElevenLabs audio APIs works against Soniqo by changing one line — the base URL. Send your key as a Bearer token or X-Api-Key. No client rewrite, no SDK migration.
Transcription is $0.24 per audio-hour — speech-to-text, diarization, and speaker ID bundled. Synthesis is $0.22 per minute of generated audio, with instant voice cloning included. Billed per second, no minimums.
Sign in with Google or GitHub for the console, or generate long-lived API keys for server-to-server traffic. Argon2-hashed, scoped to your account, revocable any time.
A durable queue with explicit acknowledgement and idempotency keys. A preemption never loses or double-bills a job — the ledger is the source of truth.
Cloud, on-prem, or on-device — same models.
Start on the managed API. Move to your own infrastructure when compliance demands it. Or run on the edge with the open-source SDKs. The models travel with you.
Managed and metered per audio-second. Sign in, get a key, and call the endpoint — nothing to host. This is what you are looking at.
Run the whole stack inside your own network. Audio and transcripts never leave your infrastructure — for regulated, air-gapped, and data-residency requirements. Custom SLAs and per-customer model tuning.
Ship speech on the edge with our open-source SDKs — speech-swift for Apple platforms, speech-core for Linux / Windows / macOS / Android — with model adaptations tuned to your hardware. No network round-trip.
One call to transcribe — one to speak.
After signing in you'll have a starter balance and one API key. Transcription is an HTTP upload; synthesis returns audio you can pipe straight to a speaker.
# Transcribe with diarization (Soniqo Steno)
curl -X POST https://api.soniqo.audio/v1/transcribe \
-H "Authorization: Bearer $SONIQO_API_KEY" \
-F "[email protected]"
# -> a job id; poll /v1/transcribe/<id> for the attributed transcript
# Synthesize speech (Soniqo Vox)
curl -X POST https://api.soniqo.audio/v1/audio/speech \
-H "Authorization: Bearer $SONIQO_API_KEY" \
-H "Content-Type: application/json" \
-d '{"input":"Hello from Soniqo.","voice":"nova"}' \
--output hello.wavAlready using the OpenAI SDK? Change the base URL and your existing code works:
from openai import OpenAI
client = OpenAI(
base_url="https://api.soniqo.audio/v1",
api_key="<your-soniqo-api-key>",
)
# Transcription
with open("meeting.wav", "rb") as f:
print(client.audio.transcriptions.create(model="whisper-1", file=f).text)
# Synthesis
client.audio.speech.create(
model="tts-1", voice="nova", input="Hello from Soniqo.",
).stream_to_file("hello.wav")Simple, per-second pricing.
Same models, same API — for free.
The cloud is for convenience. If you want full control, on-device latency, or zero per-minute cost, run the open-source library on your own hardware. Apache 2.0, nothing leaves your machines.
The same transcription, speaker separation, and voice synthesis models, packaged for Apple platforms with on-device model adaptations. Swift API, Homebrew install, Apache 2.0.
A cross-platform C++ engine for Linux, Windows, macOS, and Android — transcription, diarization, speaker ID, and realtime streaming on CPU. Apache 2.0.