> ## Documentation Index > Fetch the complete documentation index at: https://docs.prisme.ai/llms.txt > Use this file to discover all available pages before exploring further. # Talk to an agent out loud > End user: hold a spoken conversation with an agent in Chat, hear its answers, cut it off by speaking, and find the exchange in the conversation afterwards. Use voice mode when you would rather talk than type. Select **Voice mode** in the chat input, ask your question out loud and listen to the agent's answer. Speak again to cut it off. Every exchange is written in the conversation, like a typed message, so you can read it back later. ## Before you begin * **Who:** any user of Chat on the web. * **Where:** the Chat home screen, or a conversation with an agent in **Chat**. * **Requires:** * a microphone and speakers or headphones, in a recent desktop browser (Chrome, Edge, Firefox or Safari); * voice conversations available to you: your organization allows them, your role grants **Use Audio Transcriptions**, your platform administrators have set up the speech providers, and you have voice time left in your quotas. The **Voice mode** button only appears when all of this is in place. ## Talk to the agent **1. Start voice mode** In the chat input, select **Voice mode**, next to the dictation microphone. The first time, your browser asks for permission to use the microphone: allow it. * **From the Chat home screen:** a new conversation opens with the agent and model selected on the home screen, and voice mode starts in it on its own. * **From a conversation:** voice mode starts in that conversation. The chat input is replaced by the voice panel, which shows **Connecting…**, then **Listening**. The conversation stays visible above it. **2. Ask your question** Speak naturally. What you say appears in the conversation as you speak, and the panel shows **Thinking…** once you stop. **3. Listen to the answer** The panel shows **The agent is answering** and the answer is written in the conversation while it is read aloud. To cut the agent off, start speaking: playback stops at once. You can also select **Interrupt**. **4. Go on, or end the session** Ask your next question once the agent has finished. When you are done, select **End**, or press Esc while no other window or menu is open. The chat input comes back. ## Verify the result 1. In a new conversation, start voice mode and ask a short question out loud. 2. Confirm that you hear the answer and that both your question and the answer appear in the conversation. 3. Select **End**, reload the page and confirm that the conversation is listed and contains the spoken exchange. ## Control the session | Control | What it does | | - | - | | Microphone button (**Mute the microphone**) | Stops listening without ending the session. The panel shows **Microphone off**. Select it again to speak. | | **Interrupt** | Stops the answer being read. | | Speed button (**Playback speed**) | Reads the answers at 0.8×, 1×, 1.2× or 1.5×. The speed applies from the next answer and is remembered in this browser. | | **End** | Leaves voice mode and gives the chat input back. | While voice mode is on, you cannot type in the chat input. The tools, canvas and attachments you selected in the chat input do not apply to spoken turns, and the agent answers with its own model, whatever model is selected in Chat. ## Answer an approval out loud When the agent needs your approval to run a tool, the approval card appears in the conversation and the panel shows **Approval needed**. * **Say "yes" or "no".** The panel shows what it heard, and the agent goes on once your answer is taken. If your answer is not clear, the agent asks again. * **Or use the card.** Select **Approve** or **Reject** on the card, as in a typed conversation. Some approvals can only be answered on the card, such as connecting an account to a service or an approval that another person must give. The panel then shows **Answer on the card above.** ## How voice mode works * **Your conversation is saved as usual.** Each spoken turn is stored in the conversation as text, with its sources, tools and approvals, exactly like a typed message. It appears in your conversation list and stays after you reload the page. * **Pauses.** After 2 minutes without speech, the session pauses and frees the microphone: the panel shows **Voice mode paused**. The same happens when a session reaches its maximum length. Select **Resume** to go on in the same conversation. * **Network loss.** If the connection drops, the panel shows **Connection lost** and tries again on its own, in the same conversation. When it cannot reconnect, select **Resume**. * **Echo.** Your browser removes the agent's voice from what your microphone hears. If the agent stops itself as if you had spoken while you were silent, use headphones. ## For administrators A user can talk to an agent out loud when all of these are in place: | Requirement | Where | Without it | | - | - | - | | **Allow voice conversations** (on by default) | Organization › Agents Controls › Models, see [Model governance](https://docs.prisme.ai/products/ai-governance/model-governance.md#transcription-model-for-dictation) | No member of the organization sees **Voice mode**, and new voice sessions are refused | | **Use Audio Transcriptions** permission | The user's role, in the role editor | The user does not see **Voice mode** | | Speech providers | LLM Gateway configuration, see [Configure the speech providers](https://docs.prisme.ai/self-hosting/kubernetes/configuration.md#configure-the-speech-providers) | No user sees **Voice mode** | | Voice quotas | LLM Gateway configuration, see [Voice quotas](https://docs.prisme.ai/self-hosting/kubernetes/configuration.md#voice-quotas) | A user who has used up their voice time cannot start a voice session until the quota window renews | The chat checks whether voice is available to a user when the conversation opens. A setting you change applies to a user after they reload the page. ## Troubleshooting ### The Voice mode button is missing **Cause:** voice conversations are not available to you: your organization turned **Allow voice conversations** off, your role does not grant **Use Audio Transcriptions**, the speech providers are not set up, or your browser cannot record audio. **Owner:** your organization administrator, or you for the browser. **Resolution:** reload the page, then try a recent desktop Chrome, Edge, Firefox or Safari. If the button is still missing, ask your administrator whether voice conversations are allowed for your organization and whether your role grants **Use Audio Transcriptions**. ### Microphone blocked **Cause:** the browser blocked the microphone for this site, usually because the permission prompt was refused. **Owner:** you. **Resolution:** allow the microphone for the Prisme.ai site in your browser's site settings (the icon on the left of the address bar in Chrome and Edge, the crossed-out microphone in the address bar in Firefox, **Safari › Settings for this website** in Safari), then select **Try again**. ### No microphone found **Cause:** the browser found no microphone, or another application holds it. **Owner:** you. **Resolution:** connect a microphone, close any other application using it, then select **Try again**. ### Voice mode unavailable **Cause:** the platform refused the session. The panel says why: voice conversations are turned off for your organization, your voice minutes are used up, you are not allowed to talk to this agent by voice, or your sign-in session has expired. **Owner:** your organization administrator, or you for an expired session. **Resolution:** for an expired session, reload the page and sign in again. Otherwise, ask your administrator about your organization's voice settings and quotas. ### Voice mode is busy **Cause:** too many voice sessions are running on the platform, or voice is temporarily unavailable. **Owner:** none: this is temporary. **Resolution:** select **Try again** after a moment. ### Voice service unreachable **Cause:** the voice session could not start, for example because of a network filter or a temporary outage. **Owner:** you, or your IT team for a corporate network. **Resolution:** select **Try again**. If it keeps failing on a corporate network, ask your IT team to allow WebSocket connections to the Prisme.ai API. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.