Skip to content
Petal

FixesMac7 min read

VoiceInk not transcribing: diagnose the failure

VoiceInk not transcribing? Check the microphone capture, then the Mode and model, then the paste step, with VoiceInk's official diagnostics.

Published by Optimal Apps, the maker of Petal.

On this page

With VoiceInk not transcribing, the failure is in one of three steps: the microphone capture, the model step, or the paste step. VoiceInk's official "Transcription Failing or No Output" guide checks them in that order. First confirm Microphone and Accessibility in System Settings > Privacy & Security and restart VoiceInk. Then open Audio in the VoiceInk sidebar and confirm the input. Then check that the active Mode has a valid transcription model, and that the model is downloaded in AI Models. If text appears in VoiceInk but not in your app, the paste step is the problem.

Quick-fix checklist

Symptom Step that fails Fix
Recorder shows no audio level Capture Select the mic in Audio; check Sound > Input
No output after a permission change Capture Restart VoiceInk
New mic not in the list Capture Click Refresh Microphones
Mode gives no result Model Enable the Mode and pick a valid model
Local model gives nothing Model Download it in AI Models; check imported .bin files
Cloud model fails Model Check the API key and network
Text shows in VoiceInk, not in your app Paste Follow the paste guide
Text is wrong, not missing Model Try another model; add Vocabulary

Step 1: Diagnose VoiceInk not transcribing in three parts

Make one test before you change settings.

  1. Open TextEdit and click in a new document.
  2. Use your VoiceInk shortcut and say one clear sentence.
  3. Watch the recorder while you speak, then look at VoiceInk's history for the result.

Now place the failure:

  • The audio level does not move. The capture step fails. Go to Step 2.
  • The level moves, but there is no text in VoiceInk. The model step fails. Go to Step 4.
  • VoiceInk shows the text, but TextEdit is empty. The paste step fails. Go to Step 6.

Step 2: Check permissions

VoiceInk's docs list three permissions:

  • Microphone is required to record.
  • Accessibility is needed for reliable paste and for selected-text context.
  • Screen Recording is needed only when a Mode uses screen context.
  1. Open System Settings > Privacy & Security > Microphone and make sure VoiceInk is on.
  2. Open System Settings > Privacy & Security > Accessibility and make sure VoiceInk is on.
  3. Restart VoiceInk after any change. VoiceInk's docs say this step is needed.

If you built VoiceInk from source or installed a new build, the switch can show "on" and still fail. macOS ties the Accessibility grant to the app's code signature, and a new signature does not match the old grant. VoiceInk's GitHub issue #530 gives the fix: run tccutil reset Accessibility com.prakashjoshipax.VoiceInk in Terminal, open VoiceInk again, and grant Accessibility again.

For a general guide, read Mac microphone permission.

Step 3: Check the audio input

  1. In the VoiceInk sidebar, open Audio.
  2. Use Selected Microphone to pick one device. Or use Priority Order so VoiceInk falls back through several devices when one is missing.
  3. If you just connected a new device, click Refresh Microphones.
  4. Open System Settings > Sound > Input, speak, and confirm that the input level moves for the same device.

Why it works: VoiceInk records from the device that it has selected. If that device is unplugged, asleep, or muted, the model gets silence and returns nothing.

VoiceInk suggests the Mac's built-in mic or a dedicated external mic instead of a wireless headset mic. Bluetooth headsets often switch to a low-quality mode for the microphone. See Bluetooth microphone dictation on Mac.

There is also an open report (GitHub issue #916) where VoiceInk saved fully silent audio on one setup, and recorded correctly only while macOS Dictation ran at the same time. If your level meter moves in Sound settings but VoiceInk records silence, mention that issue when you report it.

Step 4: Check the active Mode

Modes decide which model and which output VoiceInk uses. A Mode with the wrong setting is a common cause of "no output".

  1. Open Modes in the sidebar.
  2. Confirm that the Mode you use is enabled.
  3. Confirm that it has a valid transcription model, and that the provider or local model is available.
  4. Confirm that the language setting matches your speech.
  5. Check the output type: Paste inserts text, Respond shows an answer in the recorder, and Custom Command sends the text to your command. If the output is Respond, no text goes into your app by design.

Step 5: Fix VoiceInk model not loading

Open AI Models and check the model that the Mode uses.

Local models

  • Confirm that the model is downloaded.
  • For an imported Whisper .bin file, confirm that the file still exists and is readable. If you moved or deleted it, import it again.
  • Streaming-only models need Real-time mode. Imported files need a batch-capable model.
  • Apple Speech needs macOS 26 or later, and the language asset installed.
  • VoiceInk Refine enhancement needs Apple silicon, at least 16 GB of RAM, and a downloaded Refine model.

On Intel Macs, VoiceInk's docs say cloud transcription usually works better, because local models are slow there. A slow model can look like no output.

Cloud providers

  • Confirm that the API key is set, and that the provider supports transcription.
  • For a custom model, check the endpoint path, the API key, and the model name. VoiceInk notes that a save does not test the connection, so a typo fails only at transcription.
  • Check the network. In GitHub issue #1010, a user behind a corporate network saw a transcription error. The cause was a cloud model selected by mistake. The fix was to download the local model and set it in the Modes.

Step 6: Separate a paste problem

If VoiceInk shows the transcription but your app stays empty, transcription works. The paste fails. VoiceInk's clipboard guide gives these steps:

  1. Confirm that the Mode uses Paste output.
  2. Click in the target field before you record.
  3. Confirm Accessibility (Step 2).
  4. Open Settings > Pasting and change Paste Method from Default to AppleScript.
  5. If old clipboard content pastes, increase the Restore Delay. The default is 2 seconds, and the range is 250 ms to 5 seconds.

The full guide is in VoiceInk not pasting.

Step 7: If the text is wrong, not missing

If VoiceInk transcribes but the text has errors:

  • Try a different model in the Mode. VoiceInk's docs suggest Parakeet V3 for fast local dictation on supported Macs.
  • Add fixed corrections in Word Replacements and important names in Vocabulary.
  • Move closer to the mic and reduce background noise.

Step 8: Collect diagnostics

If none of the steps work, use VoiceInk's Diagnostics and Support to collect system information or export logs. Review the logs before you share them, then report the problem on VoiceInk's GitHub or to its support.

Petal: a free signed build with on-device models

VoiceInk's source is GPL v3, and you can build it yourself for free. The signed, ready-to-run build is a paid one-time license. If you want a ready signed build at no cost, Petal for Mac is an option. It is free and open source (MIT), from Optimal Apps, the publisher of this site. The download on GitHub Releases is signed with a Developer ID and updates through Sparkle, so macOS permissions stay tied to the same signature between releases.

How Petal maps to the same three steps:

  • Capture. Pick the device in Settings > Recording > Microphone, or keep Default to follow the system.
  • Model. Download a speech model in Settings > Models. Parakeet 110M is the recommended light English model. Parakeet 0.6B V3 covers 25 European languages, Whisper Large V3 Turbo covers 99 languages, and Apple Speech needs no download. All run on your Mac. Petal keeps the RAM need of its default model under 8 GB.
  • Paste. Petal pastes into the front app and restores your clipboard 700 ms later, so slow Electron apps get the right text. If Accessibility is off, Petal copies the transcript instead.

Each recording and transcript is saved in History, and Transcribe Again re-runs the saved audio. Cleanup with Petal W1 runs on your Mac too. Petal needs macOS 26.2 or later.

For a side-by-side comparison, read Petal vs VoiceInk or VoiceInk alternative.

Summary

  1. Test once to find which step fails: capture, model, or paste.
  2. Confirm Microphone and Accessibility, and restart VoiceInk.
  3. After a rebuild, reset the Accessibility grant with tccutil.
  4. Select the mic in Audio and confirm the level in Sound settings.
  5. Check that the Mode is enabled, has a valid model, and uses Paste output.
  6. Confirm the model is downloaded, or that the cloud key and network work.
  7. If text shows in VoiceInk but not in your app, fix the paste step.

Questions

Why is VoiceInk not transcribing anything?

VoiceInk's docs name five common sources: a missing macOS permission, the wrong microphone, a Mode with unexpected settings, a local model too slow for the Mac, or paste timing in the target app. Check them in that order.

Why does VoiceInk show no output with a local model?

Confirm the model is downloaded in AI Models and selected in the active Mode. For an imported Whisper .bin file, confirm the file still exists and is readable. Streaming-only models need Real-time mode.

Does VoiceInk need the internet?

Not with a downloaded local model. A cloud transcription provider needs the internet and a valid API key. If transcription fails on a work network, check that the Mode does not use a cloud model by mistake.

Is there a free signed alternative to VoiceInk?

Petal for Mac is free and open source, and the release on GitHub is a signed build. It runs Parakeet, Whisper, Qwen3 ASR, and other speech models on your Mac.

FixesMac

Mac dictation wrong language: how to fix it

Mac dictation wrong language? Add your language in Keyboard > Dictation, switch it while you dictate, check the input source, and finish the download.

FixesMac

Spokenly not working on Mac: how to fix it

Spokenly not working on Mac? Fix the microphone and Accessibility permissions, the shortcut, local model downloads, Local-only mode, and paste problems.