[073]InspiringMac / Windows desktop · AI dictationSubscription or one-time purchase · direct sales on websiteData as of 2026-11-08
A push-to-talk dictation tool for Mac aimed at a narrow audience, sold by subscription or one-time purchase, capped at a few thousand a month
C
Stop building generic dictation. If you do build it, make push-to-talk a tool for one specific group, not a product in itselfThis assumes you already know that group (people who struggle to type, Chinese or Japanese IME users, clinics and law firms that can't use the cloud) and have a way to reach them. Run the math against a ceiling of roughly $1–2K a monthWorth studying · grade and headline based on the full write-up · next review 12-01
| Background | Whisper has been open source for four years, and hold-to-talk dictation has become standard. Wispr is valued at $2B |
|---|---|
| Upside | The one-person Voicy hit $20K/month (TrustMRR), but from April it flattened out. The other eight are all under $4.3K. |
| Demand | The people most willing to pay are those with typing difficulties. Bad reviews cluster around iOS keyboard switching, pricing, and Traditional Chinese / Cantonese / Japanese input |
| Competition | On Product Hunt in October: 87 similar launches, top score 167 votes; 3,000 open-source alternatives above 8 stars; Granola is free with unlimited use 10-29 |
| Revenue | Voicy October $19,975; VoiceType $4,243; the other seven $123–1,654; two listed for sale |
| Growth | Wispr has 2,864 creatives on Meta and 652 on Google. Small teams are winning with use-case SEO landing pages and YouTube tutorials. |
| Window | It's not a window, it's narrowing. Google Rambler (May), the iOS 27 system dictation (launched June) and Granola going free (10-29) keep arriving. |
| Who it's for | Only if you're already in a narrow niche and can do Mac-native or system-level input. Skip it if you just want to build a 'better Wispr'. |
Latest
- 11-08macOS 26 added SpeechAnalyzer, a local speech recognition API, lowering the bar again. The open-source project 知了 (cicada) got 1,089 stars in 4 days. It does live subtitles, hold-right-⌥ Chinese-to-foreign-language translation, and bilingual meeting notes. It monetizes by using the free tool to drive traffic to its own API credits (about ¥0.15–1 per hour of use), not by selling software. Meanwhile Handy grew from 2,268 to 37,103 stars in 7 days. It's free and open source and still climbing. Paid numbers haven't moved for Voicy (October $19,975) or VoiceType ($4,243).
- 11-07In the first days of November, two new samples appeared: the open-source "Cicada" hit 4 stars in 1,089 days with 216 forks, using macOS 26's on-device speech API for live captions and bilingual minutes, and it monetizes by having users enter their own Ling key (about ¥0.15–1 per hour of meetings). On HN, Subvocal, a silent-speech glasses startup from YC F26, scored 808 points and stayed on the front page for 29 hours. Its founder says he is a heavy Wispr Flow user but can't use it in libraries or offices. Both show that the "can't speak out loud" limitation is widely felt, but neither makes money from selling software, which matches the original's advice not to sell it as a product.
- 11-06macOS 26's on-device speech recognition API is lowering the bar for this kind of tool. The open-source Cicada built a repo in 4 days and hit 1,089 stars, offering live captions and hold-right-⌥ Chinese-to-foreign-language translation. It drives traffic with a free tool and charges through the user's own API key (about ¥0.15–1 for an hour of meetings), not by selling software. In the same period, Handy gained 2,268 stars in 7 days to reach 37,103. The free general-purpose version keeps growing, and the paid-side numbers haven't changed.
- 11-05On the free and open-source side, two more new things: macOS 26's on-device recognition API got "Cicada" to 1,089 stars in 4 days, monetized through a free tool that pushes users to a self-supplied API key (about ¥0.15–1 for an hour of meetings). YC F26's Subvocal reads mouth movements with radio waves for silent input, scored 808 points on Show HN, and its founder says he's a heavy Wispr Flow user who can't use it in libraries or offices. The paid-side evidence hasn't changed, but the scenario limits of "hold to talk" keep getting confirmed.