StrataDocs
↑↓ navigate ↵ open esc close

Voice mode

Voice mode is a hands-free, spoken conversation with Strata. You talk; Strata listens, thinks, and replies out loud.

Voice mode does not use the chat model you pick from the model picker. It runs on two specialized models your organization deploys through its Azure AI Foundry — Realtime Voice (the spoken conversation) and Realtime Transcription (turning your speech into text the AI reads). Both are entries in the Model Catalog; the voice button only appears when your organization has enabled and deployed both.

Note

Voice mode is currently best-supported on desktop Chrome and Edge. Mobile browsers and Safari work in the happy path but may run into platform-specific audio issues. The web app also includes a separate Dictate button (microphone icon) for converting your speech into text in the input box — that is push-to-talk dictation, not voice mode.

Prerequisites

  • Microphone permission for the browser tab.
  • Voice mode is hidden automatically if any of these are true: the Realtime Voice or Realtime Transcription model is not deployed in your organization's Azure AI Foundry, an admin has turned voice off in policy, or your organization is suspended. If you do not see the voice button next to Send, it is not available for your account today — and if you contact your admin, Strata's message tells them exactly which model or setting is missing.

Starting voice mode

  1. In the chat composer, click the voice mode button (the waveform icon, to the left of the Send arrow). The button is only visible when the input box is empty — if you have text typed, clear it first.
  2. Your browser prompts for microphone access. Click Allow.
  3. A full-screen voice overlay opens with a live transcript area and two controls at the bottom: Mute and End.

Strata greets you out loud. The overlay's state tells you what's happening:

  • Idle / listening — waiting for you to speak.
  • Speaking — Strata is talking, and its words stream into the transcript.
  • Processing — Strata is running a tool (for example, a SQL query) between hearing you and replying.

Speaking and being heard

Voice mode uses server-side voice activity detection (VAD), so you do not press a button to start or stop talking. Just speak. When you pause, Strata begins generating its reply.

You can interrupt Strata mid-reply by speaking — the playback stops and Strata listens again.

Mute and End

  • Mute (microphone icon at the bottom) — disables your microphone but keeps the session open. Strata cannot hear you while muted. Click again to unmute.
  • End (x icon) — closes the voice session. The full transcript is saved as a new conversation in your history (titled by Strata or the first thing you said).

Voice with a database

If a SQL Server extension is connected, voice mode uses run_sql to answer data questions out loud — it discovers schemas, runs queries, and summarizes the results verbally ("about three point two million dollars in active pipeline").

Without a database, voice mode is a general-purpose assistant — answering questions, brainstorming, summarizing — but it will not pretend to query data.

Tools available in voice

Voice mode runs nearly the same toolset as text chat. Speak naturally and Strata picks the right tool:

  • run_sql — query the connected SQL Server database.
  • search_outlook / read_outlook_email — find and read your email when Outlook is connected.
  • search_onedrive / read_onedrive_file — find and read files when OneDrive is connected.
  • search_sharepoint — search your organization's SharePoint sites when the SharePoint extension is connected (SharePoint requires OneDrive).
  • send_email — draft an email out loud, then confirm before Strata sends it.
  • generate_excel — build a styled workbook from data you've just asked about.
  • generate_pdf — build a report PDF from data or talking points.
  • generate_pptx — build a PowerPoint deck from what you've discussed.
  • save_to_onedrive — upload a generated file to your OneDrive.
  • search_web — look up live information on the public web, when web search is available on your deployment.

If a capability requires an extension you haven't connected (for example, asking for email without Outlook connected), Strata will say so and continue without it.

Files generated in voice

When Strata generates an Excel, PDF, or other file during voice, it appears as a floating preview card you can drag anywhere on screen and resize. You can:

  • Click the file card to download it or open the preview.
  • Ask Strata to email it ("send that report to my CFO") — voice routes the generated file as an attachment via send_email.
  • Ask Strata to save it to OneDrive ("save that to my Q3 folder").

A small Strata is editing badge appears while Strata composes or updates a draft (for example, an email card), and disappears once it is ready.

Action steps in the voice overlay

While Strata is thinking, the voice overlay shows the same action-step trail you see in text chat — "Querying data...", "Building report...", "Saving file to OneDrive..." — so you always know what it's doing between your sentence and its reply. If you close the voice overlay mid-task, the trail continues in the underlying chat view.

What gets saved

Voice conversations are saved to your past-chats list just like text chats. In the Recent Chats panel on the empty-chat screen, a voice conversation is labeled Voice Chat with a microphone icon, in the spot where a text chat shows the model it used. You can open one later, read the transcript and any files Strata generated, and continue in text.

Voice sessions are subject to your usage quota. Sessions also have a 30-minute hard cap — Strata closes a session automatically at the cap and tells you it ended. Strata also re-checks your sign-in roughly once a minute: if your session becomes invalid mid-call (you sign out in another tab, your role changes, or your organization is suspended), Strata closes the voice session cleanly, cancels the in-flight response, and shows a "Voice paused — sign in again" prompt.

Voice in incognito

If you start voice mode while incognito is on, the transcript is not saved when the session ends, and Strata does not see your prior text-chat messages from the same session.

Switching back to text chat

Click End. The voice overlay closes and your chat returns. If the conversation was saved, it is now in the past-chats sidebar.

What can go wrong

  • No voice button next to Send. Voice is hidden because the Realtime Voice or Realtime Transcription model isn't deployed in your organization's Foundry, an admin turned voice off in policy, or your organization is suspended. Contact your admin — the error Strata shows names the specific model or setting to fix.
  • Browser asks for microphone but Strata is silent after you grant it. The browser may have blocked audio playback. Reload the page and try again. Incognito or private-browsing windows are more prone to this.
  • Voice mode disappears mid-session. A network drop or the 30-minute cap closes the session. Click the voice button again to restart; your prior text chat is still in the past-chats sidebar.
  • Strata interrupts itself in a loop. If you are on speakerphone, the AI's own voice may bleed into the mic and trigger false interrupts. Use headphones, or speak in a quieter environment.
  • Strata says it cannot reach email or files. The matching extension (Outlook or OneDrive) is not connected for your account. Open Extensions from the input box, connect it, and try again — voice will pick up the new capability without restarting.

Related