Speech-to-Text¶

Speech-to-Text (STT) lets you dictate a chat message from the input bar instead of typing it. Lumiverse supports the browser's built-in Web Speech API and OpenAI-compatible transcription connections such as Whisper.

Setting Up STT¶

Open Settings > Voice > Speech-to-Text and choose a provider.

For an STT connection:

1. Open the Connections drawer
2. Go to STT Connections
3. Click New STT Connection
4. Enter a name, API key, and transcription model such as gpt-4o-transcribe, whisper-1, or your provider's equivalent
5. Return to Settings > Voice and select that connection

Dictating a Message¶

1. Open a chat
2. Click the microphone button in the input bar
3. Speak your message
4. Click the microphone again to finish, or use auto-submit after silence if enabled

When transcription finishes, Lumiverse places the dictated text into the chat flow.

By default, a completed STT transcript is queued as a user message. If you want Lumiverse to send it immediately and start generation, end your dictation with send message.

Auto-Submit After Silence¶

For STT connections, enable Auto-submit after silence if you want Lumiverse to stop recording automatically after you finish speaking.

This is useful for Whisper-style providers because they do not stream interim words back to the browser. Lumiverse listens for confirmed speech, then waits for a sustained pause before sending the audio to transcription.

Use this when:

• You want hands-free dictation
• Your messages were being cut off by stopping the mic too early
• You prefer Lumiverse to decide when the utterance is complete

Leave it off when:

• You want full manual control over when recording ends
• You often pause for long stretches while thinking mid-sentence
• Your microphone or room noise makes silence detection unreliable

Command Words¶

Lumiverse recognizes a small set of spoken commands while normalizing STT transcripts.

Message Action¶

send message only works as a command at the end of the transcript. If you say it in the middle, it remains part of the message text.

Formatting and Punctuation¶

Thought markers nest. The first thought start inserts *; a second nested thought inserts **. Matching thought end commands unwind that nesting.

Tips for Better Transcription¶

• Speak a little past your final word before stopping the mic manually.
• Use Auto-submit after silence for Whisper/STT connections if you frequently clip the end of messages.
• Keep the microphone close enough that speech is clearly louder than room noise.
• If auto-submit triggers too early, turn it off and stop the recording manually.
• Use send message only when you are sure the dictated message should start generation immediately.

Troubleshooting¶