Seven transcription tools I'd actually use, sorted by one question: do you run the model yourself, or pay someone else to?
Top pick WhisperX ↗︎
| # | Tool | Best for | Pricing | Rating | Provenance |
|---|---|---|---|---|---|
| 1 | WhisperX ↗︎ | Engineers who want word-level timestamps and speaker diarization in a scriptable pipeline. | Open source (self-host); compute cost only | 5/5 | researched |
| 2 | Deepgram ↗︎ | Teams shipping real-time transcription via API without managing models. | Usage-based API, free credits on signup | 4/5 | researched |
| 3 | MacWhisper ↗︎ | Mac users who want local, private transcription with a real UI and no terminal. | Free tier; one-time Pro license | 4/5 | used |
| 4 | AssemblyAI ↗︎ | Teams who want a transcription API with diarization plus extras like summarization and topic detection. | Usage-based API, free credits on signup | 4/5 | researched |
| 5 | whisper.cpp ↗︎ | Running Whisper locally on plain CPU or modest hardware, on any OS, with no Python stack to babysit. | Open source (self-host); compute cost only | 4/5 | researched |
| 6 | superwhisper ↗︎ | Mac users who want local dictation that types into any app, with models running on-device. | Free tier; paid plan for larger local models | 4/5 | researched |
| 7 | Wispr Flow ↗︎ | Talking instead of typing — system-wide voice dictation that cleans up filler and punctuation as you go. | Free tier; paid Pro plan | 4/5 | used |
Quick disclaimer: this is my own opinion, put together from what I found and tested online. The space moves fast, so treat it as a snapshot, not gospel. I'll update it when things change.
Why bother with a dedicated transcription tool at all? Because the moment you've got more than a handful of recordings to deal with, meetings, voice notes, interviews, podcast audio, doing it by hand stops being realistic. A good tool turns an hour of audio into searchable, editable text in a few minutes, and the better ones tell you who said what and where in the file they said it. That's the difference between a transcript you can actually work with and a wall of text.
One thing to get straight first: "transcription" is really two jobs. Either you're turning recorded audio into text, or you're talking and watching it type for you live. Different tools win at each. So the first question is the familiar one: do you run the model yourself, or call someone else's? The second is which of those two jobs you're actually doing.
Accuracy on timestamps and who-said-what came first. After that, the boring practical stuff: can I drop the tool into a pipeline without a fight, and where does my audio actually go? Word-error-rate I mostly stopped worrying about. Once you're past the obvious names, the good ones are close enough that it didn't decide anything for me.
If you're fine in a terminal, WhisperX is the one. Forced alignment gives it the word-level timestamps plain Whisper never had, and the whole pipeline is yours. Want that kind of accuracy without running anything yourself? Deepgram and AssemblyAI both do it over an API. Deepgram if you mostly want fast, clean transcripts; AssemblyAI if you also want summaries and topics handed back with the text.
For local transcription on a Mac, MacWhisper is what I actually use. Drop a file in, get a transcript, nothing leaves the machine. Not on a Mac, or want to go lower-level? whisper.cpp runs Whisper on plain CPU on basically any OS, with almost no dependencies. It's also quietly the engine under a lot of the friendlier apps on this list.
Then there's dictation, which is its own thing: you talk, it types into whatever app you're in. superwhisper does this on a Mac with the models running on-device, so your voice stays local. Wispr Flow does it too, and I tested it, it's fast and cleans up your ums nicely. The catch is it's cloud. Your audio goes to their servers, so know that before you wire it into everything you type.
Word timestamps and diarization in a pipeline you own. My pick if you're happy in a terminal.
Rather call an API than run a GPU? Fast, clean, low-latency transcripts.
The one I use. Local, private, real UI, no terminal.
First-hand: This is the one I actually use. Drop a file in, get a clean transcript back, done. No terminal, and the audio never leaves my Mac.
An API that also hands back summaries and topics, not just raw text.
Whisper on plain CPU, any OS, almost no dependencies. The engine under a lot of these apps.
Mac dictation that stays on-device. You talk, it types, nothing uploaded.
Cloud dictation I tested. Fast and tidy, but your voice leaves the machine.
First-hand: I tested this one for dictation. Fast, and the auto-cleanup of ums and false starts is genuinely good. Catch: it's cloud, so your voice goes to their servers.
FAQ
