How to Transcribe an Audio File Locally on Your Computer
To transcribe an audio file locally, choose a tool that opens your recording and runs a downloaded speech model on your computer. Start with a short copy of the file. Check whether the words, names, and timestamps are good enough before running the full recording.
Decide what you need from the recording
| Your next step | Output to look for | First check |
|---|---|---|
| Edit an interview or lecture | Plain text | Can you find and replay the section with an unclear name? |
| Import subtitles into a video editor | A subtitle file such as SRT | Do cues appear when the speaker starts? |
| Identify each person in a conversation | Speaker labels and timestamps | Does the tool keep speakers separate when they interrupt each other? |
If you're choosing an app, look for a local model setting, supported file formats, and the export format you need. Try a clip with a difficult name and a number. A successful export matters as much as the words on screen.
Try a desktop app if you prefer a visual editor
On Mac, open a copy of the recording in MacWhisper and choose a local model. Compare the transcript with the playback, then check the export menu for the format you need. On Windows, if you're making subtitles, open the media in Subtitle Edit and choose a downloadable local engine from its speech-to-text tools. Review the generated cues, then save a sample as SRT and open it in the app where you'll use it. Check that the format and local engine are available in your chosen build. MacWhisper transcription · Subtitle Edit's speech-to-text engines
Run a short test with whisper.cpp
If you prefer the command line, use the project's setup instructions to install or build the CLI and download a model. Save a short test copy as 16-bit WAV for the documented quick-start workflow. From the project folder, this example uses the English base model and writes plain text next to the input. Adapt the executable path to your build. whisper.cpp quick start · CLI output flags
./build/bin/whisper-cli -m models/ggml-base.en.bin -f sample.wav -otxt
Open the text file and compare a tricky sentence with the audio. If you need subtitle output, try the documented -osrt flag and open the result in your editor. For speech in another language, choose a multilingual model and check the CLI language options. Inspect speaker attribution separately before relying on a transcript to assign quotes.
Keep the original recording and a reviewed transcript
Save the transcript beside a copy of the recording. If the audio says “Mia moved the pickup to 4:15,” check both the name and the time before quoting it. If you're testing offline operation, download all required files first and run a fresh clip after disconnecting. Check the save folder if you also use cloud sync.
Voconly · Dictate a note after reviewing the recording
Once you've checked the transcript, speak your own takeaway or correction into the document with a local desktop speech model. This workflow is useful when your next step is writing.