Skip to content

01AI and integrations · module

Voice assistant and speech-to-text

Dictate queries, transcribe recordings, and work with your ERP by voice through the AI assistant

  • AI conversation
  • AI message
  • Meeting (recording transcript)
12
capabilities
3
record types
3
automations
7
integrations and AI tools

02What the module does

Less routine. More value.

The AI assistant's voice features convert speech to text and let you speak with Draivix directly. Transcription handles both chat dictation and meeting recordings, while the voice assistant replies out loud and uses the same tools as text chat to search and prepare records—so your team can work with the ERP hands-free.

  • Dictate queries instead of typingDictation in the assistant chat is converted to text by Soniox or Gemini, so you can ask questions by voice whenever you cannot type.
  • Transcripts segmented by pausesTranscription returns timestamped words and splits text at pauses longer than 1.2 s, making longer recordings much easier to navigate.
  • Recordings deleted after transcriptionUploaded files are deleted once Soniox finishes transcribing them, so audio files are not retained after transcription completes.
  • Meetings converted to textMeeting recordings are transcribed using the same drivers as dictation—Soniox or Gemini—giving you a written record of everything discussed.
  • Simply speak with the assistantThe voice assistant holds spoken conversations via OpenAI Realtime or Gemini, letting your team work with Draivix hands-free.
  • Same rules for voice and chatSearch and prepare records by voice using the exact same tools and permission checks as the text assistant. Voice follows the same rules as chat.
  • Nothing is sent without your approvalSending emails and other write operations wait until you approve them. This hold is handled by the browser's voice interface; by design, the server does not enforce a separate check.
  • Microphone right where you workAccess the microphone from global search to open a voice overlay right on top of your current page—start speaking without leaving the screen you are working on.
  • Voice quick actionsVoice quick actions are available directly on records. Trigger record-specific steps by voice without hunting through menus.
  • Time-limited voice sessionsDraivix issues an 8-hour companion token for each voice session, scoped to the user's permissions. Voice access never exceeds what the user can do in their normal work.
  • Revoke voice sessions with a single commandAn administrator can revoke voice sessions for one company or all companies with a single command—useful because logging out does not revoke the token itself.
  • Enable voice whenever you are readyThe voice assistant is off by default and enabled by an administrator. Your company decides whether and when to roll out voice control.

03How it works

How it works step by step.

  1. Speech-to-text transcription

    Speech-to-text uses the Soniox and Gemini drivers. Soniox processes audio asynchronously, returns timestamped words, and splits text into segments at pauses longer than 1.2 seconds. Uploaded audio files are deleted after transcription.

    Read in detail

    Meeting recording transcription uses the same drivers. You can dictate directly into the assistant chat or via the microphone icon in global search.

  2. Voice assistant

    The voice assistant holds spoken conversations with the user. The browser connects directly to OpenAI Realtime or Gemini. When the assistant needs data, tool calls pass through the voice bridge configured for your company and through the Draivix capability bridge—using the exact same tools and permission checks as the text assistant.

    Read in detail

    The voice assistant is disabled by default and enabled by an administrator. You can use your voice to look up a record, get a status summary, or prepare an update that you then confirm.

  3. Approval before writing data

    The voice interface holds any capability that modifies data or communicates externally—such as sending an email—until the user grants approval. This confirmation is handled by the browser's voice interface; by design, the server does not enforce a separate check.

    Read in detail

    All write operations run under the logged-in user's permissions, just like in text chat.

  4. Voice session security

    For a voice session, Draivix issues a companion token valid for 8 hours. It can only be created by an authenticated browser session, and its access is restricted to the user's permissions. Issuing a new token revokes the previous one. Because logging out does not revoke the token, administrators have a command to revoke voice sessions for a single company or across all companies at once.

  5. Where to find voice controls

    Voice controls are available on the full /chat page, in the right-hand assistant panel, via the microphone icon in global search, as a voice overlay on top of the current page, and in record quick actions. Modules can also register their own custom actions in the voice interface.

04What runs automatically

Automations that save time.

  1. Transcripts delivered automatically

    Once transcription starts via Soniox, processing runs asynchronously in the background and the result is fetched automatically—no need to wait around or monitor it.

  2. Text structured by natural speech pauses

    During transcription, text is automatically broken into segments based on pauses in speech, so you never have to wade through a single wall of text.

  3. Only the latest token stays valid

    Whenever a new voice token is issued, the previous one is automatically revoked. Older tokens can no longer be used, keeping only a single active token in circulation.

External integrations

  • Soniox converts speech to textSoniox transcribes both dictation and meeting recordings, returning timestamped words. Uploaded files are deleted after transcription so they are never retained by the service.
  • Gemini for transcription and conversationGemini serves as both a speech-to-text driver and a conversational service for the voice assistant, processing what the user says.
  • Live conversation via OpenAIDuring a conversation, the browser connects directly to OpenAI Realtime. When the assistant needs data, calls are routed through the voice bridge and capability bridge with the same permission checks.

What the AI agent handles

  • Find any record in a single sentence

    Say what you are looking for and the voice assistant finds and reads out the records—seeing only what the logged-in user is permitted to access.

  • Dictate changes, confirm them yourself

    Have the assistant prepare a record update by voice. Nothing is saved until you confirm it, and always under the logged-in user's permissions.

  • Emails sent only on your go-ahead

    The assistant can draft and send an email on a voice command, but the voice interface holds it until you approve—so no message goes out before you review it.

  • Same summaries as in text chat

    Request the same module summaries by voice as you would from the text assistant. Get an overview of sales, inventory, or production without touching the keyboard.

A person approves whenever money, a customer, or an exception is involved.

05Frequently asked questions

Can Draivix transcribe speech to text?

Yes. Transcription uses Soniox or Gemini and handles both chat dictation and meeting recordings. Soniox returns words with timestamps.

Can I talk to the ERP?

Yes, once an administrator enables the voice assistant. Conversations run via OpenAI Realtime or Gemini, and the assistant uses the same tools and permissions as text chat.

Will the voice assistant send an email without my knowledge?

No. The voice interface holds email sending and other write operations until you grant approval.

How long does a voice session last?

The companion token is valid for 8 hours and is not revoked by logging out. Issuing a new token revokes the previous one, and an administrator can revoke sessions in bulk.

Let’s start together

Tell us how your business works.

We will show you what a system that takes over part of that work could look like.

Book a consultation