Yes, you can transcribe a podcast entirely on your own device, with no upload. Get the episode as an audio file, then run it through a speech-to-text tool that keeps its model on your phone or computer. On a phone, that’s an app with on-device Whisper. On a computer, it’s a desktop app such as MacWhisper, Buzz or Vibe, or OpenAI’s free Whisper command-line tool. All of them work offline once the model is downloaded.
Here’s how to do it each way, and which one suits which kind of show.
Can you really transcribe a podcast locally? #
You can, and the results are good. OpenAI’s Whisper models are open, and projects like whisper.cpp run them on ordinary laptops and phones. Clear speech from one or two hosts transcribes well. Music beds, crosstalk and phone-quality guests make the output rougher, just as they do for cloud services.
You need three things:
- The audio as a file, such as MP3, M4A or WAV. Streaming inside an app isn’t enough.
- A local model, downloaded once. Phone models run from under 100 MB to around 575 MB. Desktop models can be larger.
- Time. Your device does the work, so a long episode on a large model takes a while.
Step 1: get the episode as an audio file #
For your own show, export the final mix from your editor. WAV or a high-bitrate MP3 is ideal. Whisper works on 16 kHz audio internally, so there’s no benefit in exporting at a huge sample rate.
For someone else’s show, most podcasts publish an RSS feed that links directly to each episode’s MP3. Some podcast apps also let you save downloads to your files. Transcribing a public episode for your own notes is common. Republishing the transcript is a different matter and needs the creator’s permission.
Step 2, option A: transcribe the podcast on your phone #
If the file is on your phone, you don’t need a computer. Private Transcribe runs Whisper on iPhone, iPad and Android and never uploads audio:
- In the Models tab, download Professional (539 MB) for a good balance on podcast audio. On recent phones with about 6 GB of RAM, Best (574 MB, Whisper large-v3-turbo) is more accurate. Use a Multilingual model if the show isn’t in English.
- Add names to custom vocabulary: hosts, guests, sponsors, show-specific terms.
- Tap Import audio and pick the episode. The app reads m4a, mp3, wav and aac, plus mp4 and mov for video podcasts. Each file can be up to 90 minutes, so split longer episodes into parts first.
- Leave the phone to work, ideally on a charger. The screen stays on while it transcribes.
- Open the transcript in History, search it, and use the Timestamps view to jump to any line.
With Pro (a one-time $7.99 in the US), you can export the transcript as .txt or as .srt subtitles with timing, which works for YouTube captions. The app doesn’t label speakers, so for a two-host show you’ll add names while editing.
Step 2, option B: use a desktop app #
Desktop apps suit bigger jobs: batches, back catalogs and shows where speaker labels matter.
- MacWhisper (Mac): drag in the episode and get an editable transcript. The one-time Pro upgrade adds batch transcription, speaker recognition and more export formats.
- Buzz (Windows, Mac, Linux): free and open source. It exports TXT, SRT and VTT and can identify speakers.
- Vibe (Windows, Mac, Linux): free and open source. It handles batches and video files, can separate speakers, and exports SRT, VTT, TXT, HTML, PDF, JSON and DOCX.
For Windows specifics, see how to transcribe podcasts offline on Windows. For Linux, see private podcast transcription on Linux distros.
Step 2, option C: use the Whisper command line #
If you’re comfortable in a terminal, OpenAI’s reference tool is free and scriptable. According to the Whisper README:
- Install ffmpeg from your package manager. Whisper needs it to read audio.
- Install Whisper with
pip install -U openai-whisper. - Run
whisper episode.mp3 --model turbo. Add--language Englishto skip language detection.
The turbo model needs roughly 6 GB of video memory, per OpenAI’s table. On a machine without a capable GPU, try small or medium. By default the tool writes the transcript in several formats, including plain text and subtitles, next to the audio file.
For lower memory use and faster CPU performance, whisper.cpp is the alternative. Build it with CMake and download a model with its download-ggml-model.sh script. Then run whisper-cli on a 16-bit WAV file. Convert MP3s first with ffmpeg -i episode.mp3 -ar 16000 -ac 1 -c:a pcm_s16le episode.wav.
Which method should you use? #
| Your situation | Best route |
|---|---|
| Episode is already on your phone, or you want the simplest option | Phone app with an on-device model |
| Two or more speakers and you want names in the transcript | MacWhisper Pro, Buzz or Vibe |
| A back catalog of dozens of episodes | Batch mode in a desktop app, or a Whisper script |
| You want full control and automation | Whisper or whisper.cpp from the command line |
How to get speaker labels for interview shows #
Plain Whisper doesn’t know who is speaking. For interview and panel shows, you have a few local options. Buzz, Vibe and MacWhisper Pro label speakers in their apps. For scripts, WhisperX adds speaker diarization using pyannote, but you’ll need a free Hugging Face token and to accept the model’s terms. Diarization isn’t perfect with overlapping voices, so check the labels by ear.
If you record each person on a separate track, you can transcribe the tracks one at a time instead. It’s more work, but you get accurate attribution.
How to clean up a podcast transcript #
Budget a pass over every transcript you plan to publish. Machine output gets most words right, but mistakes cluster where they’re most visible: guest names, brand names, numbers and jargon. Search for the names you added to your vocabulary to make sure they came out right. Listen back to anything you intend to quote. For more ways to reduce errors before you transcribe, see how to improve speech to text accuracy for podcasts.
Frequently asked questions #
Can you transcribe a podcast on an iPhone without internet? #
Yes. Save the episode to Files, import it into an app with an on-device speech model, and transcribe it in airplane mode if you like. Private Transcribe and Whisper Notes both do this on iPhone. Private Transcribe also works on Android.
How long does it take to transcribe a podcast locally? #
It depends on the model size and your hardware. Small models on a modern laptop run much faster than real time. Large models on a phone can take longer than the episode itself. If speed matters, start with a mid-size model and step up only if the accuracy isn’t good enough.
Is it legal to transcribe someone else’s podcast? #
Making a private transcript for your own notes is generally treated differently from publishing it, but copyright law varies by country. If you plan to share or publish a transcript of a show you don’t own, ask the creator first. This isn’t legal advice.
Can I transcribe a live podcast as it streams? #
Not with most private tools. Record the stream or wait for the episode file, then transcribe it. Live captioning tools exist, but most of them send audio to a server.