Offline transcription on Android: get a transcript with no connection

Using an Apple device? Follow the offline transcription guide for iPhone. This page describes the Android version of Murble.

How offline transcription works on Android

Offline transcription needs a speech model stored on your device and enough memory to run it. In Murble, a downloaded Whisper model processes recorded and imported audio locally. Once the model is installed, that processing can work without internet access.

The Android app also offers Android speech services on devices with a supported local recognizer. Availability and language support vary. For this guide, choose an installed Whisper model so your finished transcript uses the model you prepared.

An internet connection is needed to install the app and download models. It is not needed to transcribe your recordings with an installed Whisper model.

Set up recording and imported audio

1. Download a speech model

   Open Settings → Recording and transcription. Find Offline transcription and download a Whisper model. Start with Balanced for everyday recordings, or use the smaller Fastest model when storage is limited.

   Keep a connection until the app confirms the model is installed. Check the download size shown in the app and leave room for the model, your recordings and processing. Installing Murble alone does not install every speech model.

2. Choose Whisper and the recording language

   In the same screen, set Transcription engine to Whisper. This option appears after a Whisper model is installed. If you have several models, select the one you want to use.

   Choose the language you will speak. Automatic language detection is available with Whisper, but choosing a language helps when a recording is short. Turn on Transcribe automatically if you want Murble to process each new recording after you stop.

3. Record a short note

   Return to the library and start a recording. Allow microphone access if Android asks. Speak for about thirty seconds, then stop and let Murble produce the finished transcript.

   The finished transcript is the result to check first. A live preview during recording is a separate feature and may need additional setup on your device.

4. Try an existing audio file

   Use Import audio file from the library to bring in a recording you already have. For the offline test, choose a file saved on your device. A file that exists only in a cloud storage app must be downloaded before you disconnect.

   Murble transcribes the imported audio locally using the installed model. Processing time depends on the length and quality of the recording and the performance of your phone or tablet.

5. Read, correct and export

   Open the transcript, review names and numbers, and correct recognition mistakes. Time aligned lines let you return to the relevant part of the recording.

   Export a local copy when you need it. Generating a transcript or a local export does not require Google Drive sync. Sending that file by email or saving it to a cloud service needs whatever connection the destination requires.

Which model size to pick

• Fastest: short, clear recordings when storage and speed matter most.
• Balanced: everyday recordings, with a balance of speed and accuracy.
• Best accuracy: more difficult audio when your device has room and time for a larger model.

Download sizes are approximate. The app shows the current size before you download. A larger model also needs more working memory. Try a smaller model if your device struggles with processing; no model guarantees a perfect transcript.

What about live transcription?

The words shown while you speak are a preview. Murble uses a separate live recognition flow, with available Google ML Kit speech recognition, local Android speech services and installed Whisper as alternatives. The engine available to you depends on your device, language and installed models.

To try it, turn on Live transcription in Settings → Recording and transcription, choose your language and make a short recording while connected first. Let any required speech model or language download finish, then repeat the test offline.

If the live preview is unavailable, you can still record and get a finished transcript afterwards using your installed Whisper model. A working preview in one language does not prove that every other language is ready offline.

Test it in airplane mode

1. Finish installing your chosen Whisper model and select it before disconnecting.
2. Turn on airplane mode and check that Wi-Fi is also off. Some Android devices remember a previous Wi-Fi setting in airplane mode.
3. Record thirty seconds, stop and open the finished transcript.
4. Import a short audio file that is already saved locally and check its transcript too.

If either test fails, reconnect and check that the model finished installing, that Whisper is selected and that the file is available on the device. Repeat with a smaller model if memory is the problem. For a fuller inspection, see how to verify offline processing.

Limitations and privacy

Sources

• Murble for Android: model choices, downloads and engine selection in Settings → Recording and transcription, checked 8 October 2026.
• Android documentation: availability of a local speech recognizer.
• Google ML Kit speech recognition: model setup and supported languages and devices.
• Murble Android privacy policy and support: optional sync, diagnostics and troubleshooting.

Keep reading