🎤Micdrop

Speech to Text Providers for a Voice Agent

The speech to text transcribes the voice of the user, utterance by utterance, and hands each transcript to the agent. It is passed to the server as stt, and on its own it turns the server into a dictation tool.

Hosted

  • Gladia, fast and accurate multilingual transcription
  • Gemini Transcribe, streaming transcription in 85+ languages
  • OpenAI, real-time transcription with gpt-live-transcribe
  • Mistral Voxtral, real-time transcription with Voxtral 3 in 40+ languages
  • Gradium, real-time transcription on an EU endpoint

The hosted ones differ in price, languages and time to the final transcript. Providers without a ready integration, such as Deepgram or AssemblyAI, appear in a ranking of streaming speech to text APIs for voice agents.

Local

  • Whisper, transcription inside your Node process, with no API key

The local speech to text guide helps pick a model.

Comparison

ProviderLatencyLanguagesCost per hourBest For
Gladia 🇫🇷~200ms99+$0.75General use, multilingual
Gradium 🇫🇷~300ms5$0.62French, EU data residency
OpenAI Whisper~300ms57$0.36High accuracy, multilingual
Whisper (local)~440ms99FreePrivacy, no API key

Costs are the published list price of the entry paid plan in August 2026, per hour of audio transcribed. Every provider charges less as the volume grows, and Gladia goes furthest with real-time transcription at $0.25 an hour on a commitment.

Your own

Any other service plugs in by extending the STT class, and FallbackSTT switches to another provider when one fails.