Speech (STT & TTS)

Pia includes local speech processing — no cloud APIs required.

Speech-to-Text (STT)

Pia runs speech-to-text locally on your machine and offers two engines you can choose between:

• Parakeet (the default) — multilingual, and it detects the spoken language automatically.
• Whisper — lets you pick a model size to balance speed and accuracy.

Choosing an engine

1. Open Settings → General → Speech
2. Pick your engine from the Speech-to-text engine dropdown
3. Click the Download button to grab the required files — you only need to do this once per engine

Parakeet

Parakeet is the default engine. There’s no model size to choose — it works out of the box and detects the language you’re speaking on its own. Just click Download to set it up.

Whisper

With Whisper, choose a model size from the dropdown: Tiny, Base, Small, Medium, or Large. Then click Download to download that model.

Tip

Start with the Base model for a good balance of speed and accuracy. Larger models are more accurate but slower.

Recording

Once your engine is set up, click the microphone icon in the chat to start recording.

Text-to-Speech (TTS)

Pia synthesizes speech in the app itself, offline, and fetches its voices from Pia’s own servers. There is no separate speech engine to download, and networks that blocked the third-party model host Pia used before can install a voice again.

Downloading a Voice

1. Open Settings → General → Speech and scroll to Text-to-Speech
2. Scroll through the voice list — each card shows the voice name, language, gender, and quality
3. Find a voice you like and click Download
4. Wait for the progress bar to finish

You can download as many voices as you want. Alba (English GB) is the voice Pia reads aloud in out of the box.

Selecting a Voice

A voice is selected as soon as its download finishes, so speech works without a second step. The card shows an Active badge to confirm which one is in use.

Only one voice can be active at a time. To switch, click Select on a different downloaded voice. If the saved voice is no longer on this device, Pia takes up one that is rather than staying silent.

Removing a Voice

Remove from this device on a voice’s row deletes it and gives back the space it took. Pia asks first — “Remove the voice X from this device? You can download it again at any time.” — and if it was the voice in use, Pia moves to another installed one.

Two voices are no longer offered

The recordings Lessac and Ryan were built from are licensed for research and non-commercial use only, so Pia no longer offers them. If you had one selected, Pia moves to another installed voice or asks you to pick one.

Note

Voices downloaded by versions of Pia before the in-app speech engine cannot be used by it and are removed on first start, which gives back the space they took. Choose a voice again under Settings → General → Speech.

Using TTS

Once a voice is active, AI responses are read aloud automatically. Spoken answers begin sooner than they used to, and the short phrases Pia says while it is thinking are ready as soon as a voice is installed.

Voice Conversation

Instead of typing and reading, you can just talk. Click the Voice conversation button in the message bar and Pia opens a full-screen overlay: it listens, thinks, answers out loud, then listens again. It uses whatever engine and voice you set up above.

How a turn works

1. Listening — speak normally. A level meter shows Pia is hearing you.
2. About a second and a half of silence ends your turn automatically, so you don’t have to press anything.
3. Processing — Pia transcribes what you said and works out a reply.
4. Speaking — the reply is read aloud, and appears in the running transcript on screen.
5. Back to listening, until you leave.

Three buttons sit at the bottom:

Button	When it shows	What it does
Done	While listening	Ends your turn now, without waiting for the silence.
Stop	While speaking	Cuts the reply short and starts listening again.
End	Always	Leaves voice mode.

Everything you say and everything Pia answers is added to the chat you were in, so you can scroll back through it afterwards like any other conversation. Replies are attributed to the active persona, same as typed ones.

Tools in voice mode

There’s no way to show you an approval card when you’re not looking at the screen, so voice mode is stricter than the chat window. A tool that changes something runs only if it’s already covered:

• you marked it Always allow, or
• Agent autonomy is on and it’s one of Pia’s own write tools

Anything else is refused out loud, and Pia tells you to ask again in the chat window.

Note

An Allow this session grant does not carry into voice mode — only a standing Always allow does. See Tool Permissions.

Privacy

All speech processing happens locally on your device. Audio is never sent to external servers.