Speech-to-text APIs

What it is

A speech-to-text API accepts audio and returns a transcript, often with timestamps, punctuation and speaker labels. Streaming APIs return text while someone is still talking. Batch APIs process recordings.

How it works

• Streaming: open a WebSocket, send small audio chunks, and receive interim and final transcripts. Deepgram’s Flux model also emits turn events for voice agents (see turn detection).
• Batch: upload a file or URL and get the full transcript back.
• Accuracy helpers: formatting (punctuation, numerals) and keyterm prompting to bias recognition toward names and jargon.
• Buy instead of build: Wispr Flow is system-wide dictation that cleans up your speech as it types, in any app. Pro is $15 per user per month billed monthly, and a developer API is available to approved organizations (checked 2026-10-17).
• Local alternatives: open models such as Whisper run on your own hardware for privacy-sensitive cases.

Pricing (checked 2026-10-17). Deepgram gives $200 of free credit, then bills per minute. Streaming Nova-3 is listed at $0.0048 a minute and Flux English at $0.0065 a minute on pay as you go. Both are promotional rates, with regular prices of $0.0077 a minute. Budget on the regular rate.

Benefits

• Hands-free input (dictation) and searchable records of calls and meetings.
• The foundation for voice agents. See the Jarvis build guide.

Limits and risks

• Consent and disclosure. With cloud STT, audio leaves the device. Tell people and get consent to record.
• Accuracy varies. It drops with noise, accents and jargon, so test on your own audio.
• Promotional prices end. Per-minute rates vary by model and plan, so re-check before you budget.