Birmingham solo founder Ibrahim Farah built the desktop app around local Whisper transcription, editable subtitles and one-click Anki exports.
By RuntimeWire Staff · Published
Primary source: SubSmith
Why it matters #
SubSmith shows how solo founders can build paid products above open-source AI models by owning a narrow workflow. Farah's bet is that local processing, editing and Anki export are worth paying for even when transcription itself is becoming a commodity.
Birmingham developer Ibrahim Farah put SubSmith in front of a new group of early users on August 28, offering language learners a desktop app that turns their own video and audio files into editable transcripts, vocabulary prompts and Anki flashcards.
The push follows months of public testing. SubSmith appeared on AlternativeTo on January 22 and was featured by BetaList on February 17. Its changelog shows Farah was still shipping setup and diagnostic improvements with version 1.3.4 on August 14. The latest post is another distribution push for an existing product, rather than SubSmith's first release.
Farah built SubSmith during evenings and weekends, describing it in a LinkedIn post as his first desktop-app project. The idea came from his own trouble studying podcasts, YouTube videos and lectures that lacked usable transcripts. He wanted to feed in material he already had, generate subtitles and stay inside one study workflow instead of moving among a media player, dictionary and flashcard software.
SubSmith is operated by Farah as a sole trader in Birmingham, according to its terms of service. That solo structure helps explain the product's narrow pitch and low pricing: $15 a year or $40 for a lifetime licence, following a seven-day trial that does not require a credit card.
Whisper handles the transcript; SubSmith handles the study session
SubSmith accepts local media including MKV, MP4, MP3 and WAV files. It uses OpenAI's Whisper for speech recognition and FFmpeg for media processing, according to the product's About page. Farah has built the application around the steps that follow transcription: correcting words and timestamps, hovering over vocabulary for definitions, replaying a line and sending the sentence to Anki with its audio clip and an optional screenshot.
SubSmith says it supports more than 99 languages through Whisper. That figure describes the underlying model's language coverage, rather than a separate curriculum or teaching system developed by Farah. Transcription quality will still vary with the language, recording quality, background noise and model performance. SubSmith's terms explicitly stop short of guaranteeing error-free transcripts.
Farah also chose local processing as a product constraint. SubSmith says transcription runs on the user's computer and that media files, transcripts and subtitles remain on the device. Its privacy policy says the application sends limited diagnostic data, including a random installation identifier, app version and operating-system family, while excluding filenames and media content.
That design fits the material SubSmith is targeting. Up a movie or a folder of podcast recordings to a cloud transcription service can be slow and expensive, while a local model can process the files without transferring the underlying media. It also moves the computing cost to the customer, keeping the listed annual price at $15.
The $40 lifetime option comes with a careful qualification. The licence itself does not expire, but updates, online services and compatibility depend on how long SubSmith remains supported. For a solo-operated desktop application, that clause matters: operating-system updates, Whisper changes and third-party integrations can create recurring maintenance work long after the initial sale.
The competition already knows about local files
Farah is entering a well-developed category. Language Reactor built its audience around interactive subtitles on Netflix and YouTube, while Migaku can create flashcards from video subtitles with dictionary entries, screenshots and audio clips.
SubSmith's dedicated desktop workflow is a useful distinction, though local media alone no longer separates it from every incumbent. Language Reactor has a media-file player that can use embedded subtitles, import subtitle files or generate text from audio with speech recognition.
Farah is therefore competing on focus and packaging. SubSmith combines local Whisper transcription, timestamp editing, word lookup and direct Anki exports in a native application priced for individual learners. Language Reactor and Migaku carry broader sets of learning and content-discovery features, which can make them stronger general-purpose products and busier tools for someone who mainly wants to process downloaded files.
The underlying transcription technology is available to any developer. Whisper and FFmpeg are open-source components, and general-purpose AI tools continue to make transcription and translation cheaper. SubSmith's defensibility rests on whether Farah can make the editing and sentence-mining workflow dependable enough that learners pay to avoid assembling the same process themselves.
A small product built around a specific frustration
SubSmith reflects a familiar path for solo software founders: start with an open-source technical capability, wrap it around a personal pain point and charge for the tedious integration work. Farah does not need to train a speech model or build a licensed video catalog. He needs SubSmith to open a learner's files reliably, produce a transcript worth correcting and move useful lines into Anki with fewer clicks than the alternatives.
The pricing keeps that promise testable. A seven-day trial gives users enough time to run their own accents, recording conditions and target languages through Whisper before paying. It also leaves Farah little room for expensive support or cloud inference, making local processing part of the business model as well as the privacy pitch.
SubSmith's homepage calls the product "More effective than a class," a marketing comparison that is not substantiated by evidence in the product materials. The more credible pitch is narrower: learners with folders of downloaded media can turn those files into searchable, editable study material without up them or wiring together several utilities.
That is a modest promise, and it gives Farah a concrete product to improve. The next test is whether enough language learners share his frustration to turn an evenings-and-weekends desktop project into durable paid software.