Soniox

Soniox is a speech-to-text API built around a unified model that handles transcription, translation, diarization, and timestamps in a single stream — all priced together rather than as add-ons. The differentiating capability is true code-switching: if a speaker flips between English, Spanish, and French in one sentence, Soniox transcribes the mix correctly without you pre-selecting a language.

60+ languages are supported. Real-time streaming returns tokens within milliseconds, which is critical for voice AI agents and live captioning. Pricing is token-based but works out to roughly $0.10/hour for async file processing and $0.12/hour for real-time streaming — competitive with raw Whisper API and lower than Gladia's Starter tier.

A consumer-facing Soniox app exists at $19.99/mo Pro for unlimited transcription, translation, summaries, and insights, but the product is clearly API-first and the consumer experience hasn't reached Otter or Notta polish levels. The cons: token-based pricing takes math to predict (input audio tokens, input text tokens, output text tokens — all priced separately), the developer community is smaller than around Whisper, and the documentation is solid but assumes engineering competence.

Best for developers building voice products with multilingual or code-switching audio, AI meeting tools targeting global markets, contact centers with mixed-language traffic. Wrong fit for non-technical users wanting a finished transcription UI — pick Happy Scribe or Sonix instead.