The Athena Guide · 03

Voice

The three microphone modes, Athena's voice, video narration, the phone

The three microphone modes

Chosen in Settings → Voice and appearance, per device:

• Hold to talk: you speak while holding the microphone, and when you let go the question goes out. Pressing it while Athena speaks stops her; while she works (thinks or uses tools) what you say becomes a direction.
• Open: one tap turns it on and it stays on. It listens even while she speaks, so you can interrupt her by talking, but a noise or another voice can stop her too.
• Closed until she finishes: one tap turns it on; it does not listen while she thinks and speaks, then reopens by itself. To stop her use ■ or Esc.

Athena's voice

She usually speaks with the engine's voice. If the engine does not provide it or does not answer, she uses the system voice (the phone's or the computer's). You can choose it in Settings → Voice and appearance. A written answer can be heard with read, under the message.

She speaks the language you chose in Account → Language: the engine's voice speaks ten (Italian, English, Spanish, French, German, Portuguese, Russian, Chinese, Japanese, Korean), with the same timbre; for other languages she uses the system voice in that language.

A voice for videos

Under Settings → Connections → Other AI → OpenAI: images and voice, enable Use OpenAI to generate audio. Reuse your saved personal key and select a TTS model and voice. Athena chooses reading instructions per segment, based on the scene and your requests, while keeping the narration coherent. OpenAI becomes the default for genera_voce; you can explicitly request the local engine instead. Disabling the option restores the engine default. Text is sent to OpenAI and API charges apply. Image key sharing does not grant speech access to other users.

Check TTS model verifies model access without generating audio. This setting controls narration files for editing, separately from reading chat replies aloud. Lore follows the configured permissions; other modes request confirmation for OpenAI synthesis.

This uses the Speech API /audio/speech. OpenAI documentation checked on October 24, 2026 announces retirement of its current TTS models on January 6, 2027. Migrating to Realtime requires a separate adapter, rather than replacing the model name. Disclose that the voice is AI generated when sharing the video.

You can ask her: “Create the narration for this video and add it to the images.” When the connected engine provides speech synthesis, Athena uses it through genera_voce, without installing another program to create the voice.

The result is a WAV file, saved in audio/ in the chat's working folder and attached to the reply. Athena measures its duration and can use it with FilmCraft for video editing and SoundCraft for audio mixing.

For a new video with narration, create the voice first: its duration guides the clip lengths and editing. The current engine accepts up to 4096 characters per segment; longer narration is prepared in several parts, using the voices available in the engine.

What counts as speech

When holding the microphone you don't need a question: «Hi Athena», «thanks» or a direction arrive as they are. Only sounds and noises («uhm», «mm») are dropped. Her name at the start («Athena, what time is it?») is removed.

On the phone

• Voice works only on a secure page (https): the address your administrator gave you already is.
• After an Athena update, close and reopen the browser tab.
• On iPhone, with the screen locked or in another app, Safari stops listening: for a meeting the page must stay open (the screen stays on by itself).