Documentation

Changelog

Recent changes to the Soniqo Cloud API — new endpoints, capabilities, and compatibility.

2026-08-24

Breaking: language is now required on transcription

Every transcription request must name its language. Pass language on /v1/audio/transcriptions, inside the JSON config part on /v1/transcribe, and as a query parameter on the realtime upgrade. A request without one now returns 400 instead of being transcribed with a guessed language.
Why. Requests that arrived without a language were sent to a classifier that listened to the audio and picked a model. On held-out recordings it sent roughly four in ten Arabic-English clips to the European model, which transcribes them very poorly — and its own accuracy check had been measured on the same recordings it learned from, so the failure never showed up. Rather than rebuild the classifier, we removed it: the caller almost always knows the language, and a stated language is always right.
What to change. Add the language you already know to each call. If you genuinely do not know it, detect it once on a short sample with any language-identification tool and pass the result — the accuracy will be better than the guess this endpoint used to make.

2026-07-24

Voice cloning is no longer available in Soniqo Cloud. The /v1/voices and /v1/text-to-speech/* routes now return 404.
Preset speech synthesis remains available through /v1/synthesize and the OpenAI-compatible /v1/audio/speech endpoint. Use /v1/models/synthesize to list built-in voices.
Previously registered reference data is no longer used for synthesis. Contact support if you want it deleted.

2026-06-23

Text-to-speech — generate natural speech via /v1/audio/speech (OpenAI-compatible) and /v1/synthesize.
Voice cloning (retired 2026-07-24) — this release originally allowed registering a short clip with /v1/voices and reusing it on synthesis calls. It was included with no per-voice fee.
Low-latency streaming — 24 kHz PCM from response_format=pcm, and Authorization: Bearer now accepts an API key, so OpenAI SDKs and agent frameworks like LiveKit work with a base-URL change. See the voice-agents guide.
Console — synthesis from the browser.

2026-06-23

Transcription $0.24 / audio-hour, synthesis $0.22 / minute, billed per second.
Free tier is now $10 / month of credit — no card needed to start.

2026-06

Speaker diarization and identification bundled into every transcription.
Expanded language and dialect coverage; auto-detect or hint the language.
OpenAI-Whisper-compatible (/v1/audio/transcriptions), async for long audio (/v1/transcribe), and realtime streaming (/v1/realtime).