The Archive · Android · Guides

Dictate, listen, and use your own voice

Speak instead of typing, have replies read aloud, and train a personal voice from twelve phrases — all on the phone.

Checked on Android 0.1.38

A newer version is out (Android 0.1.40): some details may differ.

Everything about speech is in Settings → Voice, apart from the interface language. To open TALOS by voice, see Call TALOS from anywhere.

Tap the microphone in the composer and speak. Your voice becomes text on the phone; the audio is never stored or sent. Dictation needs the Microphone permission, asked the first time.

Dictation language is Automatic by default: TALOS works out which language you speak, and follows it even mid-sentence. You can also follow the phone’s language, or pick one.

Tap the speaker on any reply to hear it. Under Read replies aloud, choose the reading voice — a voice of the phone, by default its own — and its rate and pitch, then Preview voice. Turn on Show voices for every language to choose beyond your own.

TALOS can read replies in your voice, built from twelve phrases you record once. Recording, analysis and speech all happen on the phone.

  1. Under Personal voice, tap Download the voice engine — about 160 MB, once.
  2. Tap Create your voice and confirm the three statements: that you are the person being recorded, that the file stays encrypted on the phone and can be deleted, and that the microphone is used only while you record.
  3. Find a quiet place: hold the phone about 20–30 cm from your face, and speak in a natural tone.
  4. Record the 12 phrases — four whispered, four normal, four loud — holding the button for each. A phrase too quiet, too loud, too short or too silent is refused with the reason; record it again. It takes about 3 minutes.
  5. The whole set is checked together, then Encode the voice.
  6. Listen to how it turned out. If it does not convince you, train it again; otherwise name it and Save voice.

The recordings are encrypted with a key held in the phone’s hardware keystore, and the raw audio is destroyed once the voice is built: what remains is the voice, not the recording.

Each voice can be previewed, renamed or deleted, and one is in use for reading. Delete forever deletes its encryption key too: the file becomes unreadable for good, even if a copy remained. A voice marked Incompatible was made by an older engine and needs to be recorded again.

Type to search the guides.