Skip to content
MakerGain
← Back to cases

MacWhisper Turns Recordings Into Text, Then Leaves Identifying the Speakers to You

Edited by MakerGainPublished: Article updated:

Podfeet's Allison uses MacWhisper for interview captions and transcripts: she enters a 12.6-second offset, exports an SRT, then labels speakers sentence by sentence and merges paragraphs. Automatic recognition can separate voices, but it cannot decide identities for her; she uses ChatGPT to help with segmentation and runs into confusion between an API key and a Plus subscription. Developer Jordi Bruin tightens the terms because of license resale.

Making captions for an interview video usually takes more than one mouse click. Allison of the Podfeet blog wrote about how she and Steve divide the work: Steve finishes editing the video, exports the audio as uncompressed AIFF to her desktop, and appends to the filename the offset between the start of the video and when she starts speaking—for example, 12-600 means 12.6 seconds. After transcribing in MacWhisper, she selects the segments, enters that offset, exports an SRT, and gives it to Steve to upload to YouTube. For a 3 to 7 minute video, transcription takes one or two minutes; entering the offset and exporting takes only a few seconds. [1]

MacWhisper is a transcription app for the Mac. Its website says lectures, videos, podcasts, and private files can all be dragged in; it supports more than 100 languages and exports, can record online meetings without a bot joining, and can dictate in any app. MacWhisper[2] turns sound into text; the model takes over this step. The trouble comes afterward: even if the machine splits the voices into Speaker 1 and Speaker 2 for you, a person still has to turn those into real names like Franci and Paweł. Allison's case was more primitive—they had recorded only mono, so she could not take the Podcast separate-track automatic speaker recognition route; she had to identify the speaker manually for every sentence.

Captions are the easy part; the transcript is what takes effort. What Allison wants is not a timeline with one sentence per line but whole paragraphs of each person speaking. The app has a one-click way to merge segments into sentences, but merging them into paragraphs still has to be done by hand; and before merging, every sentence first has to have a name. She hovers over a sentence and presses '1' or '2' to identify the speaker, pressing while playing the audio; if she clicks slightly off, she enters edit mode and types '2' into the middle of someone else's sentence. To merge paragraphs, she holds shift to select several and then right-clicks merge—MacWhisper developer Jordi Bruin promised her a shortcut for merging, but she had not gotten it yet when she wrote this. [1]

Later, on another Mac, she saw a green button labeled Separate Speakers. It uses MacWhisper's ChatGPT feature, not the built-in automatic speaker recognition. Pressing it hit an error: You do not have access to model 'GPT-4o'. The error message said her OpenAI account needed a valid payment method and that she first had to spend at least $1 on GPT-4o mini. She pays $20 a month for ChatGPT Plus and thought that money would be enough. Jordi replied explaining that the ChatGPT feature requires an API key, which is not the same thing—he said he knew it was confusing and had written to OpenAI. Allison also said she had lots of questions, and Jordi and Ian responded quickly on support. [1]

She authorized a minimum of $5 for this key and canceled recurring payment because she did not know how much it would end up costing. Back in MacWhisper, she dragged in the audio, wrote the speaker names into the prompt, asked for the names to be bolded and the result laid out as Markdown, and pressed the green button again; this time it worked. Copying could not be done with ordinary paste—she had to click copy reply on the right, and what came out was decent Markdown. [1]

This boundary is worth pausing at. The official documentation states that MacWhisper's ChatGPT feature requires users to enter their own key, and that after it runs the user reviews and applies the reply. How to use ChatGPT in MacWhisper[3] The website describes AI enhancement as custom prompts, local and cloud models, and letting users use their preferred AI services. MacWhisper[2] People who want to keep recordings on their own machine can run local models; to use cloud transcription, they can connect ElevenLabs or Deepgram; to use AI enhancement such as ChatGPT, they need to enter their own key. The optional Assistant subscription puts cloud transcription, AI translation, and 'chat with the transcript content' on the developer's servers; those who do not want this layer of hosting can switch to their own key. MacWhisper & Whisper Transcription Difference[4] Making that change may also run into errors such as quotas, model permissions, and payment methods.

Before she discovered that button, Allison had also tried a workaround: have MacWhisper export plain text and give it to ChatGPT to segment. She first tried Franci's interview, then Paweł's with the Polish accent; both times it split them correctly. Paweł's was 3.5 minutes; ChatGPT took 1 minute 45 seconds, and she estimated that doing it manually would take at least 15 minutes. She also noticed that MacWhisper merged Paweł's repeated starts into clean sentences and deleted her own 'um' and 'ah.' When she switched to Tucker Jones's interview, ChatGPT put Tucker's name where hers should have been. Fancy Transcripts of Video Interviews - Podfeet Podcasts[1] This route assumes there is already an initial transcript—ChatGPT cannot run Whisper in that environment. What is saved is post-processing; the entry point is still in MacWhisper's hands.

What is being discussed here is the automatic speaker recognition built into MacWhisper 12, which is not the same path as the ChatGPT button above. It has boundaries of its own. The official documentation says it works only with local models on the WhisperKit engine, or with the cloud transcription providers ElevenLabs and Deepgram; dictation and 'Transcribe Podcast' are not supported. Automatic Speaker Recognition in MacWhisper[5] After recognition is complete, the app assigns Speaker 1 and Speaker 2 by default; users can rename them, merge two segments spoken by the same person, and adjust segment assignment with number shortcuts. Automatic Speaker Recognition in MacWhisper[5] The machine can separate several voices, but it cannot decide for you who they are.

Image: Source Good Snooze / MacWhisper[5], accessed 2026-11-14; original interface retained.

How this business charges also determines what it looks like. The direct-download version goes through Gumroad: a one-time lifetime personal license covering all features, and users pay only once. The App Store version is called Whisper Transcription and uses weekly/monthly/yearly subscriptions; it cannot accept an external license key, and because of store restrictions, the AI provider options are fewer. MacWhisper & Whisper Transcription Difference[4] MacWhisper license keys that had been sold later became objects of resale.

A documentation update in December 2025 laid this out. The developer said he had discovered people reselling MacWhisper licenses, especially on Chinese platforms such as Taobao, so he began enforcing the terms more strictly: a personal license is for use on one's own devices, with a soft limit of 3 devices, and the system automatically disables it when it detects abuse. “Your license is disabled” error when activating[6] He listed four consequences: buyers think they have bought a legitimate license; he does not receive license revenue but still receives support emails; the system automatically bans based on signals; and every user has ongoing costs, so he is losing money on every user with a resold license. He did not give an amount, so the loss can only stop at that qualitative statement; public materials also have no revenue or profit figures.

MacWhisper comes from independent developer Jordi Bruin. Jordi Bruin - Slices: The Deep Dish Swift Podcast[7] What is interesting is how he patches the hole. He said he is working on a better licensing system, but for privacy reasons he wants to know as little as possible about where people are and how they use the app. “Your license is disabled” error when activating[6] The more closely you check, the more you can block resale—and the closer you get to the kind of app he does not want. A tool that keeps recordings local ultimately has to choose between 'checking clearly' and 'not asking much': a personal license has a soft limit of 3 devices, but he says he does not want to know where people are or how they use it.

Continue reading

AI audio
The Source Code Is Out There; the Installed Version Sells for $25: VoiceInk Puts Dictation in Any Text Field

VoiceInk is an instant-dictation app for Apple Silicon Macs: press a shortcut to write speech into a text field, or display an AI answer or run a command. Its source code is public under GPLv3; the prebuilt version is a one-time purchase at $25, $39, or $49, licensed by device count and including updates and support; cloud services require your own API key. Its affiliate program pays 30% commission but bans paid traffic, and the pricing page and README disagree on the macOS requirement.

AI videoAI audio
Video Stays on the Local Device, Only the Audio Track Is Sent Out: AutoSubtitles' $3,254 MRR

AutoSubtitles puts video-to-subtitle work into a web page: registration-free use; the page itself states that the video file stays on the local device and only the audio track is sent for transcription and deleted afterward; TrustMRR shows 21,169 US dollars in all-time revenue, 3,254 US dollars MRR, and 414 active subscriptions, with revenue verified by Stripe. The page lists Free, Pro at 9 US dollars/month, and Max at 15 US dollars/month, but does not state the billing period or the limits for each tier.

AI companionsAI imagesAI audio
One reply counts as one, a photo as eight: Embraces.AI breaks companionship into messages and sells it

Embraces.AI prices companionship by message: one reply counts as 1 message, and one Picture this image counts as 8. Companions have cross-session memory, send multiple bubbles in a row, and initiate contact; users can edit memories and export chats. Before signing up, there are 40 free messages; the three paid tiers start at $12, and $5 tops up 250 messages. TrustMRR recorded on 2026-11-14 all-time revenue of $48,193 and 426 active subscriptions.

AI imagesAI videoAI audio
59 Million Views but Almost No Revenue: After Publishing the Entire Project, Overseas Signups Increased to 6.5x on the Day

Zohar Vanunu used his self-built Kolbo.AI to make a “biblical phone imagery” AI series, which drew 59 million views and more than 200,000 followers in 30 days, but by his own account it brought in almost no revenue; after publishing the complete production tutorial, overseas signups increased to 6.5x on the day, and in three weeks 3,285 people accounted for 78% of new signups. The product puts more than 100 models, along with chat, image, video, music, and voice, into the same project and charges by credits.