Voice in Poneva Chat

The current voice features are available for review in a controlled integration build. They are not a general managed rollout. This page describes the behavior being reviewed and the safeguards that apply to it.

Dictation and voice mode

Dictation puts your words into the composer as editable text. You can change or discard the text before sending it. Voice mode is a separate, optional hands-free experience: tap to enter or leave the mode. When you finish an utterance, a short undo window gives you time to edit or discard it before it is sent. There is no hold-to-speak mode.

Language detection is automatic. Your recent messages in the current chat can help it recognize the language you are speaking. If it gets this wrong, tell Poneva—for example, “I am speaking Swedish, not Norwegian.” A correction about your speech is separate from asking for a reply or the app interface in Swedish.

You can ask Poneva to remember your usual spoken language. It proposes a personal setting for you to approve; a correction alone does not save one. You can also ask it to clear the default. This is a hint, so switching languages remains possible. Company languages and the Swedish/English pilot defaults supply fallback hints. Meetings use their own language context.

Dictation and meeting captions aim to preserve the language and script actually spoken, including language switches. Translated alternatives returned alongside the source words are excluded. Speech recognition can still make mistakes, particularly with quiet pauses and nonverbal acknowledgements.

There is no required language selection or voice picker. Review dictated text before sending, especially short phrases or names. Voice mode appears only when your company has enabled it and spoken replies are available. A deliberate action is required before the microphone is used.

Drafts, edits, open review cards, a hidden page, or a degraded connection can keep a spoken message from sending automatically. Always ask keeps the message for your approval. Voice mode releases the microphone when you leave it.

Spoken replies and interruption

Use Read aloud on a written answer to hear its complete wording. Inline code, such as a commit identifier or a command name, stays in its sentence. Completed answers are spoken together where possible; long answers continue in larger passages. Read aloud does not automatically summarize or rewrite the answer. Code blocks and tables remain on screen with a short spoken cue in the nearest identifiable language from the surrounding answer. An uncertain passage does not prevent a cue when another nearby passage identifies its language. You can stop or replay the answer, and playback stops when you leave its chat or account.

When spoken replies are available, Chat can read a short, source-linked part of an answer while the complete answer remains visible. Tables, comparisons, results, and drafts remain available in the visual workspace instead of being read row by row. You can stop playback with the control or by tapping the transcript or waveform.

During an interview, the spoken question and written recap should ask the same question. Chat waits for an answer to a confirmation question before moving to the next interview question. If an answer arrives without its spoken part, Chat can complete the speech while keeping the written answer intact.

Speaking while a reply is playing can stop that playback and continue with your next message. The visible answer remains intact and shows an interruption notice. This interruption behavior is still subject to physical speaker, headphone, and device acceptance in the review process.

Searching by voice

Web search, page reading and merchant lookups from Poneva's tool catalog run without an approval card, just as they do in typed chat. Each tool still needs to be currently authorized to run.

Confirmation and settings

Voice instructions use the same confirmation rules as typed instructions. In an ordinary personal Chat, asking Poneva to create or revise a private Artifact lets it save that work without another Save or Build tap. It reports saved only after the writer confirms the saved version. A requested new private Result can also save in that Chat loop; replacements and protected turns may still require review. Personal settings, model changes, sharing and connected actions keep their confirmation rules. The assistant cannot approve its own proposal. Knowledge changes requested in voice mode keep their applicable review controls and current publishing-right checks. A pending page suggestion does not change the approved text.

Use the confirmation card to approve or cancel a proposed change. Substantial work opens in the visual workspace for review and editing. Where spoken approval is available and enabled in both company and personal settings, your next clear voice message can approve one eligible reversible change just shown to you. Unclear acknowledgements leave it pending. High-risk changes and connected services still require the approval control. Suggesting that a knowledge page be archived always requires that control, and only a publisher's tap on the page accepts a suggestion; a typed, dictated or spoken message cannot do either.

Voice settings such as the undo window and wait before sending are personal settings that persist after reload. Changes requested through chat follow the confirmation rules. Undo restores the prior setting where the action supports it.

Recording meetings

Choose what to record before starting: your microphone, the meeting audio, or both. Browser support for meeting audio varies. A browser that offers call-tab audio asks you to select the tab and share its sound; Poneva saves audio only. Where that is unavailable, an exposed computer-audio input or an updated Poneva desktop app can provide another capture path. The computer-audio setup links to desktop downloads. Successful input choices are remembered and checked again before reuse; a missing input needs a new selection. On an updated desktop app, the initial choice includes computer audio and your microphone.

The desktop review build can remind you when another application uses the microphone. This is an optional indication that a call may be happening. The reminder appears in the center of the app over a dimmed, blurred backdrop. Choose Not now, close it, or press Escape to dismiss it. Taking notes opens the recording choices; it does not start recording for you.

Before you start, Poneva shows whether live transcription is available. If it is unavailable, the saved recording is still transcribed after you stop. Separate microphone and meeting-audio levels help you check both sides. If the audio engine cannot start, Poneva releases the inputs so you can try again. These paths remain subject to the device acceptance described below.

Speaker recognition in meetings

Speaker recognition is optional company data. An owner enables it, and each person explicitly consents before enrolling their own voice. Recognition is limited to the company and meeting context. Medium-confidence matches remain suggestions; stronger matches show their source and can be undone.

You can withdraw your enrollment. Turning recognition off, leaving the company, or deleting the account removes the stored enrollment and open suggestions. Enrollment creates an encrypted voice template and keeps limited source metadata; it does not make the original recording a reusable template. Company administrators see counts and audit information rather than another person's template. Strong automatic matches require the current consent and calibration checks; less certain matches remain suggestions.

Meeting editors can review bounded voice samples, merge or separate labels, and undo a change. Readers can see the meeting record but cannot change its speaker labels. If the voice pass is unavailable, the transcript remains usable.

Review status

The review build has automated and controlled local evidence for these flows. Final integration review, physical device acceptance, and any managed rollout are separate decisions. No statement on this page should be read as production availability or as a claim of completed speaker-accuracy or echo-cancellation acceptance.