Dialectic Speech-to-Text Services
Speech-to-text converts microphone recordings into player input. The same text then follows the normal targeting, player subtitle, player voice, dialogue, and interruption flow.
Recommended
Parakeet is the default Quickstart selection. Deepgram is the recommended hosted alternative.
Parakeet
Parakeet runs locally through its DwemerDistro component and requires no speech API key. Select Parakeet in Quickstart or Configuration - STT, then test it from the connector page.
Deepgram
Deepgram provides fast hosted transcription. Create an API key, save it in DialecticServer, select Deepgram, and test the microphone path before launching Fallout.
Other Services
OpenAI Whisper
The Whisper connector sends recorded audio to the configured hosted Whisper-compatible endpoint using its API key.
Local Whisper
Local Whisper uses a locally hosted transcription service. This avoids per-request speech charges but uses local CPU or GPU resources.
Gemini
Gemini can transcribe recorded player audio through Google's service when the connector and API key are configured.
Azure
Azure Speech requires a valid subscription key, region, and endpoint configuration.
Inworld
Inworld speech-to-text uses the credentials and endpoint configured for the Inworld service.
Microphone Controls
Bind Voice Chat in MCM for push-to-talk. Open Mic, microphone device, mute state, sensitivity, end delay, and the Open Mic Mute hotkey are configured from the Sound MCM page.
Push-to-talk begins recording on key press and finishes when released. Empty or failed transcription should not send a dialogue request.
Troubleshooting
- Confirm Windows allows Fallout: New Vegas to access the microphone.
- Check that MCM shows the intended recording device.
- Test the STT connector in DialecticServer.
- Check the game plugin and server STT logs for recording, upload, transcription, and response timings.