Skip to content
SpeechRouter

Every speech model.
One API.

Route speech-to-text across 12 providers with one key, one schema, and mid-stream failover. Switch vendors by editing a slug. Pay vendor list price — we add nothing.

Quickstart →Browse the models
12
providers
36
priced models
0%
markup
live
api.speechrouter.ai

Thirty seconds to a transcript

Section titled “Thirty seconds to a transcript”
Terminal window
curl -s https://api.speechrouter.ai/v1/audio/transcriptions \
  -H "Authorization: Bearer $SPEECHROUTER_API_KEY" \
  -F model=deepgram/nova-3 \
  -F file=@meeting.wav

Change deepgram/nova-3 to soniox/stt-rt-v5, openai/gpt-4o-transcribe, or any of the live models — nothing else in your code moves.

One schema

Every provider speaks the same events: transcript, speech_started, utterance_end, done. Words, timestamps, and speakers are normalized — switching vendors never rewrites your parser.

Mid-stream failover

Give listen() a fallback lane. If a provider dies mid-session, the stream fails over live — timestamps stay monotonic and a provider_switched event tells you it happened.

Honest pricing

Billed at vendor list price with 0% markup, in each vendor’s native unit. Bring your own provider keys and pay nothing at all.

Deepgram drop-in

Already on the Deepgram SDK? Point it at api.speechrouter.ai and keep your code — the gateway speaks the Deepgram dialect natively.