Push-to-Talk

Push-to-talk is the quickest way to speak to Atlas. Click the microphone button, talk, click again when you’re done. Atlas transcribes your words and sends them as your message — no typing required.

How It Works

1. Click the microphone button in the chat panel
2. Speak your message clearly
3. Click again when you’re done
4. Atlas transcribes your audio
5. The transcribed text appears in the chat so you can see exactly what was sent

The whole process usually takes just a second or two for short messages. Longer recordings take a little more time to process.

What Happens to Your Audio

Your audio is transcribed on your own device by Whisper Base, the local speech-to-text model Atlas downloads on first launch. Nothing is uploaded — the recording never leaves your machine, and transcription works even with networking disabled. The transcribed text becomes your chat message, and the audio is discarded once it has been transcribed.

Whisper Base transcribes English. Speak clearly and it handles most accents and everyday vocabulary well.

Tips for Good Transcription

Tip

A quality microphone makes a noticeable difference in transcription accuracy. Even the built-in microphone on a laptop works well in quiet conditions, but an external or headset microphone is often better in noisy environments.

• Speak at a normal pace — you don’t need to slow down or enunciate artificially
• Reduce background noise when possible — music, fans, and open windows can affect accuracy
• Start speaking right away — there’s no countdown, so begin when you wish
• Keep messages focused — shorter, clearer statements transcribe more reliably than long run-on sentences

Correcting Transcription Errors

If the transcription isn’t quite right, you can send a follow-up message to clarify. Atlas understands context, so a quick “I meant X, not Y” works fine.

Tip

Push-to-talk pairs with text-to-speech for a hands-free loop that works fully offline. Atlas speaks its replies with the on-device Kokoro voice; installing the optional espeak-ng system package sharpens pronunciation of unusual words and names.