Voice

ArxDeck supports optional voice features in chat and markdown composers: read aloud, dictation, and conversational mode (chat only).

Organization speech-to-text

Your organization admin chooses the speech-to-text (STT) provider at Organization → Voice:

• Browser — free, uses your browser's built-in recognition (may be unreliable on mobile Chrome).
• Deepgram or AssemblyAI — paid streaming recognition with more reliable continuous dictation.

The provider applies to all projects in the organization. You cannot override it in account settings.

Organization admins can also choose an optional Deepgram or AssemblyAI model. When no org model is selected, the platform default for that provider is used. For Deepgram, admins pick a model family first and then a specific variant when multiple streaming models share that family.

Before your first paid-provider session, you must accept a one-time consent acknowledging that audio is processed by a third-party provider. Audio streams directly from your browser to the provider; ArxDeck does not store raw audio.

Account settings

Under Account settings → Voice (/settings/account), configure text-to-speech speed, pitch, voice, and conversational mode behavior (send mode, assistant-turn handling). Account settings also show which STT provider your organization uses and your consent status.

Read aloud

Rendered markdown supports full-document read aloud, section playback (icon beside each heading), and selection playback (select text in the body, then choose Read selection). A floating bar at the bottom provides pause, resume, and stop while speech is active.

On mobile, read-aloud supports Stop only — pause and resume are not available.

Composer Preview includes section and selection controls; the Write tab textarea does not.

Agent input requests

When an agent asks structured questions on a ticket, scheduled task run, or in chat, use Read aloud on the message or comment for the introductory text, and Answer by voice to hear each question and dictate answers hands-free. The voice wizard reads questions only—not the intro. Answers appear in the visual form; you can edit or submit manually at any time. Answer advancement follows your account Voice send settings (wait-for-command vs auto-continue after idle). In chat with conversational mode on, the intro is read once when an input request arrives, then the wizard starts on question 1 automatically.

Conversational mode (chat)

When enabled in a chat thread, the app listens continuously until you end the mode. You can speak commands such as send, sent, wait, cancel, or stop listening / end conversation / stop conversation to end the mode, depending on your send settings.

Costs

Paid STT usage is recorded per session. Project admins can view speech-to-text totals on the project Settings → Costs page.

AssemblyAI session duration is rounded up to whole seconds before billing. Platform administrators configure each AssemblyAI model's USD/hour rate on Platform → Voice.

Related

• Chat & proposals
• Tickets — structured input and Answer by voice
• Organizations — org Voice settings