Voice conversation in the web chat: you speak, the assistant answers aloud

1How to start a conversation

1. Open the main chat and sign in

   The voice conversation works in the assistant chat at gibrid.ai/chat and only for signed-in accounts.

2. Click the headphones button

   It is at the top right of the chat header, next to the Temporary chat button. On hover the caption is “Voice conversation”. A panel appears at the bottom of the screen.

3. Look at the panel before you start

   It says “Speak — I will answer by voice” and shows the price: “≈ N G per minute, the standard voice is free”. The Start and End buttons are there too.

4. Click “Start” and allow the microphone

   The browser shows a microphone access request. If you refuse, the panel says “No microphone access”: allow access in the site settings of your browser and click “Start” again.

5. Speak

   Panel statuses: “Listening…”, “Hearing you…”, “Thinking…”, “Answering — you can interrupt”. When you are done, click End: the microphone turns off and the panel closes.

2What happens between your phrase and the reply

• The pause. The microphone listens all the time. Recording begins when you start speaking, and after about 0.9 seconds of silence the phrase counts as finished. One phrase lasts no longer than 30 seconds: after that it is sent anyway.
• Recognition. The recording goes to speech recognition, the same route as the voice input button. If the text isn't recognized or the recording is too short, the panel simply keeps listening.
• The model's reply. The recognized text goes into this same chat as an ordinary message. In Auto mode a fast model answers (if no file is attached), in the conversational call style: one to three short phrases, with no lists, headings, links or code. If you chose a model by hand, that model answers.
• Reading aloud. When the reply is fully ready, it is read in the standard call voice (up to 1,500 characters; code, links and formatting are not read). This voice is free.
• Interrupting. Speak while the assistant is answering: the voice stops at once and the panel listens to you.
• The feed stays. Every message and reply is visible in the chat as text, and images and documents appear as cards, as in a normal conversation.

3What a minute of conversation costs

The number on the panel is calculated in advance. It includes about three replies of the fast model per minute and speech recognition. Reading aloud in the standard voice costs nothing. The exact number depends on your plan: the lower the plan's markup, the smaller the amount.

4What you need, and how to speak so everything works

• Sign-in and balance. The conversation isn't available to guests; speech recognition needs a positive balance.
• Microphone and browser. You need a modern browser that can record sound and permission for the microphone for gibrid.ai.
• Headphones. It is better to speak with headphones: if the sound of the reply gets into the microphone, the panel may take it for your speech and interrupt the reply.
• Quiet around you. The start of speech is detected by loudness, so loud noise nearby can start a recording for nothing.
• Don't pause for long in the middle of a phrase. A second of silence is the end of the phrase: recognition starts without waiting for the rest.

5What the voice conversation on the website doesn't have

• There is no choice of voice: the assistant answers with one standard free voice.
• The reply is read aloud after it is fully ready, not as it is typed, so there is a delay between your phrase and the voice.
• Code, links, lists and tables are not read aloud; you can read them in the chat feed.
• The conversation works only in the main assistant chat. Chats with souls, groups and the “Models” section have no such button.
• It is not a call: there are no “Interrupt” and “Mute” buttons as in the app; you interrupt by voice.

6How it differs from a call in the app

More on calls with a choice of voice on the page about calls in the app.

Related