Troubleshooting

Troubleshooting

A model will not download

• Verify internet access and available disk space.
• Keep Voice2Win open until the download finishes.
• Choose a different model-storage folder or the alternative AI-model folder when the current drive is full.
• Delete an incomplete model in Manage Models and retry.

A model will not load or the computer becomes slow

• Choose a smaller or more strongly quantized model.
• Check the combined memory requirement of speech recognition and AI Enhancement.
• Assign one workload to the main processor and the other to the graphics card.
• Update the graphics driver so Vulkan or DirectML is available.
• If a remote server is selected, check its address, port, model name, and connection state.

The microphone is missing or silent

• Confirm operating-system microphone permission for Voice2Win.
• Select the intended Recording Device again and speak while watching the level indicator.
• Make sure the microphone is not muted in the operating system or on the device.
• Reconnect a removed Universal Serial Bus (USB) microphone and wait for automatic reinitialization.
• If no local microphone exists, use the Mobile Device Remote Control or a Companion Client.

Text is not inserted

• Place the cursor in an editable field before starting the recording.
• On Windows, if the target application runs with elevated rights, run Voice2Win elevated as well. The Task Scheduler autostart option can do this automatically.
• Try Clipboard insertion first, then Unicode or Scan Code.
• Increase Keyboard simulation delay when characters are missing or out of order.
• For remote desktop and terminal tools, try another paste shortcut or Scan Code mode.
• Do not change the active window or cursor position while transcription or AI replacement is running.

Clipboard insertion changes or loses clipboard contents

• Keep Restore previous clipboard enabled.
• Increase Wait time before insertion if the target pastes old content.
• Increase Wait time before restoring if the target pastes nothing or incomplete text.
• Other clipboard-monitoring applications can see the temporary content during insertion; use Unicode or Scan Code if this is unacceptable.

Recognition says “No speech detected”

• Speak closer to the microphone and reduce background noise.
• Lower Speech detection sensitivity so quieter speech is accepted.
• Reduce Clean up recording start if very short recordings lose their first word.
• Select the correct fixed transcription language.
• Add important names and domain terms under Custom specialist terms and names.

Global shortcuts do not work

• Ensure the shortcut is defined and not assigned to another Voice2Win action.
• Check whether Conversation Recording is active; it temporarily disables other shortcuts.
• Some system-reserved combinations cannot be captured reliably; choose another combination.
• On Linux, use an X11 session. Pure Wayland sessions can block global key state and text injection.
• In a sandboxed macOS build, operating-system restrictions can hide features that require global automation.

Real-time transcription is unavailable

• Block Transcription requires the main audio workload to use suitable graphics-card acceleration and enough graphics memory.
• Stream Transcription requires a supported language and its separate model download.
• On Windows, DirectML must be available for graphics-card stream transcription; select the main processor when it is not.
• Use Restore Defaults after aggressive block-parameter tuning.

Artificial intelligence (AI) Processing and Replace does not work

• Enable AI Enhancement and download or select a working model.
• Select editable text before pressing the replacement shortcut.
• Verify that the target supports normal copy and paste shortcuts.
• Keep the selection and cursor unchanged until replacement completes.
• Make the spoken instruction longer than the configured minimum transcription length.

Mobile device cannot connect

• Scan the current QR code again; a rotating security link changes after restart.
• In local mode, put both devices on the same Wi-Fi and temporarily disable mobile data.
• Choose the correct network adapter and open the firewall port when prompted.
• For phone microphone use, install and trust the generated CA certificate, reopen the secure address, and allow browser microphone access.
• In Tailscale Funnel mode, verify Tailscale installation, sign-in, Funnel approval, and internet access on both devices.
• Use Rebuild Connection from the connection-help panel.

Companion client cannot connect

• Confirm Client Connections are enabled on the host.
• Copy a fresh token-protected connection address.
• Verify that the selected host adapter and client are on the same network.
• Allow the configured port through the host firewall.
• Enable at least microphone or shortcut transfer in the client.

Still stuck?

Collect the exact status or error text, Voice2Win version, operating system, selected model, and relevant hardware assignment, then open a GitHub issue. Do not attach private recordings, access tokens, passwords, or full log files without reviewing them first.