At a glance
- Wake word
- “Hey Herald”, recognised on the phone. “Hey Hermes” still works.
- Transcription
- Your Hermes speech-to-text (
sttinconfig.yaml) - Spoken replies
- Your Hermes text-to-speech in the bot’s voice, with the phone’s voice as a fallback
- Where
- One-to-one bot chats on iPhone, iPad and Android 13 or later
- Defaults
- Hands-free off, Read replies aloud on, Listen with the screen off off
Three ways to use your voice
| Mode | How you start it | What the bot receives |
|---|---|---|
| Voice note | Tap the mic, talk, tap Send | The transcript; you keep a voice bubble with the words underneath |
| Dictation | Tap the mic, talk, choose Convert to text | Nothing until you edit the text and send it |
| Hands-free | Say “Hey Herald” in an open chat, or choose Talk hands-free from the chat menu | Each spoken turn as text, with replies read aloud |
How a hands-free conversation works
When you say “Hey Herald”, Herald starts recording, and a short pause ends your turn. The clip goes to your Hermes, which transcribes it with the speech-to-text provider you configured, such as local Whisper, Groq or OpenAI. The bot receives only the words, so it costs the same tokens as typing.
When the reply finishes, Herald reads it aloud through your Hermes text-to-speech in the bot’s own voice, as long as Read replies aloud is on, then listens again. Tap while it speaks to interrupt. Two empty recordings in a row, Stop, leaving the chat or losing the connection end the conversation. Calls and Siri pause it, and it carries on afterwards unless the interruption lasted more than ten minutes.
A conversation you started keeps going with the phone locked. Listening for the wake word with the screen off is a separate opt-in, described below.
Turn it on
- Check speech-to-text on your Hermes
Voice needs
stt.enabledand a provider. If either is missing, Herald says “Speech-to-text is turned off” or “No speech-to-text provider” and points you tohermes setup› Voice on the computer. - Switch on Hands-free
Open Settings › General › Voice › Hands-free. It then reads “Listening for ‘Hey Herald’ while a chat is open”.
- Allow the microphone and speech recognition
Herald asks the first time. The wake word needs both.
- Decide on screen-off listening
Listen with the screen off keeps the wake word active while the phone is locked or Herald is in the background. It is off by default.
- Choose whether replies are spoken
Read replies aloud is on by default. Switch it off to keep a conversation going and read the answers instead.
Privacy and battery
The wake word only ever runs on your phone’s on-device recognizer. If your phone has none for English, Herald shows the wake word as unavailable rather than stream the microphone to a server. Nothing reaches your Hermes until you say the wake word, and recordings go only to your own Hermes.
Listen with the screen off keeps the microphone on in the background. iOS shows its microphone indicator, Android shows a “Herald is listening” notification with a Stop button, and the battery drains noticeably faster.
Limits
- The wake phrase is English and needs on-device English recognition. On Android it needs version 13 or later and the offline English speech model.
- Hands-free runs in one-to-one bot chats, not group rooms. Voice notes work in rooms.
- Herald doesn’t listen while you type, while a bot is mid-turn or while CarPlay is talking.
- It is separate from Hermes’s own “Hey Hermes” wake word, which listens on your computer’s microphone.
Questions
Is this the same as the wake word in Hermes Desktop?
No. Hermes has its own wake word for the terminal and desktop app, which listens on the computer. Herald’s listens on your phone and needs no wake_word setting on Hermes.
Do voice notes cost more than typing?
No. Your Hermes transcribes the clip and the bot receives only the text, so the model sees the same words you would have typed.
Whose voice reads the replies?
The bot’s, from the text-to-speech settings in its Hermes profile. The Hermes default provider, Edge TTS, needs no key. If speech fails, the phone’s own voice takes over.
Can Siri start a conversation?
Yes, on iOS 18 or later. “Hey Siri, talk to Herald” opens your voice bot in a hands-free conversation. See Siri and Control Center.
Why does Herald say “Wake word paused”?
The listener failed several times in a row and stopped retrying. Tap the mic to talk, or switch Hands-free off and on again.
