# Voice input

Turn microphone recordings into editable text and review it before sending.

Updated: 2026-11-02
Desktop Mac production client passed this manual acceptance round; Windows microphone unverified · Web Production speech API verified; physical browser microphone unverified
Canonical: https://ai.z-group.cn/docs/en/voice-input/

The production speech service is enabled. rc.75 beta provides Mac Apple Silicon and Windows x64 packages, with Android rc.75-mobile.1 distributed under the existing internal compatibility certificate. The final Android APK passed an Android 10 physical upgrade, microphone denial/regrant and a full 60-second recording/transcription; drafts and attachments remain and nothing is sent automatically. A Mac production client previously passed physical voice tests. Real-account voice in the final integrated desktop package, the Windows microphone and a physical browser microphone remain unverified. Previous results do not cover every platform. Older packages retain their build-time offline guide; see online release records for current availability.

## Requirements

You need a working microphone, recording permission, an internet connection and an enabled speech service. After signing in, the voice window shows the provider and your remaining daily recording allowance. The current budget is 600 seconds per account per day, with a 60-second limit per recording; the dialog shows the live remaining allowance. When the service is unavailable, continue typing or use system dictation. This version has no in-app local speech recognition settings.

## Record and review

1. Open the intended conversation and click **Voice input** next to the composer.
2. Read the provider notice, consent to cloud transcription for this recording and click **Start recording**. Allow microphone access when prompted.
3. Speak a complete request and click **Stop and transcribe**. A recording ends automatically after 60 seconds.
4. Review and edit the transcript, especially brands, model numbers, quantities, dimensions and units.
5. Click **Insert into draft**. Your existing draft remains available and you can continue editing.
6. Press Send when satisfied. Finishing a recording, receiving a transcript and inserting text do not execute a task.

For example: “Find a matte, light beige stone for a hotel lobby, within three hundred yuan per square metre.” Noise and accents can affect recognition, so verify dimensions and prices yourself.

## Cancellation and permissions

Cancel or close the window to discard the transcript. Changing conversations, signing out or moving the app into the background invalidates the current result so it cannot enter another conversation. An already submitted cloud request may still finish and count toward provider usage. MATLOOK does not store original recordings; the provider's data policy applies. Inserted text follows normal draft and conversation behavior.

If permission was denied, enable microphone access in your operating system's application privacy settings. Web clients also need website permission. If no sound is detected, check the selected input device, headset connection and whether another app is occupying the microphone.

## Network, allowance and execution computer

A connection failure, timeout or exhausted daily allowance preserves your existing text draft. Continue typing as needed. Sending a task from a phone still requires its execution computer to be online with synchronization enabled. Speech recognition does not replace that computer. Finish any active Chinese input-method composition before inserting the transcript.

This Mac and Android acceptance round covered draft preservation, cancellation, manual edits and consecutive insertions, permission recovery, background interruption, and transcription network failure and recovery. Android also passed the full 60-second automatic-stop and transcription retest. Windows microphones, headsets and every network environment remain outside this result. Industry homophones may still be incorrect; review material names, brands and dimensions.

This version retains per-recording consent and manual sending. A request already sent before cancellation may still consume provider usage; text input remains available.

Microphone permission is checked only when you start recording. Opening a conversation or restoring sign-in does not request recording permission.

When voice service is unavailable, the dialog shows safe recovery instructions without internal interfaces or raw service errors. Your existing text draft is preserved.