Deepgram Nova-3 The best STT provider depends on your languages, audio conditions, latency needs, formatting expectations, and data policy. Record a representative sample, compare the transcripts, and consult the current documentation and billing terms for Deepgram, Groq Whisper, OpenAI Whisper, AssemblyAI, and any other candidate. You can switch providers without losing the rest of your settings.
Verified provider setup
TL;DR configuration Checked against OpenTypeless v1.1.48 on 2026-07-16.
OpenTypeless setup summary for stt-deepgram Provider route Deepgram Nova-3 streaming speech recognition Settings area Settings → Speech Recognition Credential Deepgram API key Transport Built-in WebSocket streaming route Built-in model nova-3
Built-in speech recognition reference Copy built-in STT reference In Settings → Speech Recognition, select deepgram and enter the provider API key. The model, endpoint, and extra fields below are built into OpenTypeless v1.1.48 and are read-only for this provider setup.
config_mode=built_in_stt
select_provider=deepgram
enter_api_key=provider_api_key
built_in_model=nova-3
built_in_endpoint=wss://api.deepgram.com/v1/listen
Observable success check With provider deepgram selected in Settings → Speech Recognition, click Test. Success is the green connection result with a latency value after OpenTypeless validates the saved key with Deepgram.
Troubleshooting The Test button stays disabled. Likely cause: No Deepgram API key is available for provider deepgram.
Fix: Enter the Deepgram key in Speech Recognition settings and run Test.
Test reports an authentication failure. Likely cause: The saved key is not accepted by the built-in Deepgram provider route.
Fix: Re-enter the key for deepgram and retry with the currently documented provider route.
The connection test passes but dictation does not stream. Likely cause: The network can validate the key but cannot open the Deepgram WebSocket route.
Fix: Allow outbound secure WebSocket access to api.deepgram.com and try a short dictation again.
Get API Key Create a Deepgram account and open project dashboard. Create an API key with speech-to-text permissions. Copy and store the key securely. Configure in OpenTypeless Settings > Speech Recognition > Provider: Deepgram. Paste API key. Keep language on Auto Detect unless a fixed language is required. Click Test and verify latency is returned. Recommended Configuration OpenTypeless uses Deepgram live WebSocket mode with Nova-3 and smart formatting.
Provider: deepgram
Protocol: WebSocket
Endpoint: wss://api.deepgram.com/v1/listen
Auth header: Authorization: Token <DEEPGRAM_API_KEY>
Model: nova-3
Audio: linear16, 16000 Hz, monoOperational Tip 💡 If multilingual sessions show unstable punctuation, set a fixed language first and compare before reverting to auto detect.
Related documentation