Voice and audio
Dictation, voice notes, live translation, audio mastering and AI copies of a person's voice and advice.
Voice and audio holds 22 of the 252 stories on Aiolum: 15 run by a solo founder and 7 by a team of 2 to 10. Together they report US$8M in verified outcomes; the largest is Wave AI, US$4M annual recurring revenue, November 2024. 11 headline numbers carry evidence grade A (disclosed or verified), 2 grade B (independent press) and 4 grade C (founder claim), and 5 are grade D (reported, not verified). The median story reached its headline number 16 months after launch. The founders are in 16 countries, most often United Kingdom and Not stated in the sources. The most common main channel is paid ads (5 of 22). The tools named most often in their sources are Apple App Store (14), Stripe (14) and Next.js (11).
Verified, number by number
Evidence gradesA disclosed or payment-verifiedB independent pressC the founder's own figuresD reported, not verifiedHow we verify
Filter: team, grade, outcome and more
More filters: main channel, size and time
The revenue trend compares the latest months in the sources with the months before them. Falling means revenue is about 15% or more below the best recent months, or the sources report a fall. The headline number is the yearly figure or the sale value on each card.
22 of 252 stories shown
It reached the App Store early, in July 2023, with the simplest form of the idea: press one button, get a transcript and a summary. The founder paid for Meta ads while they still worked for a new kind of app, tested paywalls constantly and moved buyers to annual plans. He works alone with AI coding tools, so in his own 2026 account about US$3 million of about US$7 million a year stays as profit.
Read the storyThe first apps that combined OpenAI's Whisper and ChatGPT for voice notes appeared in 2023, and the founder found them "not implemented well enough". His team built a simpler one, collected early users and feature requests through lifetime deals on AppSumo in 2024, and in 2025 grew with advertising and paywall tests measured in Adapty's subscription data.
Read the storyFor its first 20 months the product brought in about US$3,500 a month. Then three things came together: phone apps with weekly plans from August 2025, web plans with unlimited use at US$4 a month billed yearly, and a new group of customers, people who make songs with AI generators such as Suno and need the last step done for them. Monthly revenue rose 7.2 times between January and July 2026. No source states which of the changes caused the rise.
Read the storyIt started as a free Chrome extension that the founder built for his father, then sold unlimited dictation everywhere for a low monthly price. The AI model is cheap to rent: Groq lists Whisper at US$0.04 to US$0.111 per hour of audio. Users came from Google: the site has more than 1,000 pages, one for each app and each need, and the founder says that an AI agent he made writes the blog posts.
Read the storyChurches with members from many countries meet every week, and the older solutions cost too much for most of them: an interpreter for each language, or radio receivers to hand out and collect. Glossa charges US$5 for one hour of speech in one language, counts by the minute, and charges nothing for silence, songs or extra listeners, so a small church can start with four free hours and then a few dollars a week. The product is shaped for this one buyer: captions go into ProPresenter and OBS, which churches already run, and a church can fix how names, Bible quotations and its own terms are translated. A very small team can offer this because the speech recognition and the voices are rented from model companies and paid by use.
Read the storyIt reached people inside the chat app that they already open all day, so capturing an idea on a phone took one voice note. It aimed first at Notion users, whom the company told that ideas get lost on the phone. Ads on Facebook and Google then paid for fast growth in early 2026, until the founder stopped them in August 2026 because the costs were larger than the revenue.
Read the storySpeech recognition became cheap and accurate enough for one person to transcribe millions of podcast episodes. The founder sells the result to people who need to know what was said: market researchers, advertisers, analysts and journalists. The headline number comes from Stripe through TrustMRR: US$5,003 of revenue in September 2026.
Read the storyIt sells a strong need, understanding speech one cannot follow, at a high price: US$34.99 a month or US$199 a year. The hard parts, recognising speech and translating it, are rented from AI providers, so the seller works on the app around them. The seller names word of mouth, App Store optimization and Apple Ads as the channels, and has changed the store name and texts several times.
Read the storyIt answered a request that people already typed in public: what does the person in this viral video say? A free bot replied inside X, where the request was made, and the website sold the same service for longer files at a few dollars a month. The speech and translation models are bought from other companies, so one person could build and run it.
Read the storyIt packages rented music models from Google, reached through the model host fal.ai, as one studio with a clear credit price, a commercial licence and an AI label on every file. TrustMRR, reading the company's Stripe account, showed US$2,686 in monthly recurring revenue and 154 active subscriptions on 25 October 2026, five weeks after the founding date on the profile.
Read the storyIt serves a group that the large caption tools serve badly: speakers of a language with little training data who also switch languages inside one sentence. The founder already published AI lessons in Armenian, and the About page says his audience asked for this tool. Stripe data on TrustMRR shows revenue rising every month from June to October 2026.
Read the storyA marketer who had learned to code one year earlier built it in a week, found that paid ads could buy customers for about US$50 each with a yearly plan that paid the ad bill up front, and sold the business for cash eleven months after launch, when the bugs and the support load became more than one person wanted to carry.
Read the storyAcquire.com wrote that it began as a free tool to collect email addresses for the founder's affiliate marketing business, and that the paid version grew from upsells to those sign-ups. It was early in AI voice-overs while the voices were improving fast, and the price rose as they did: from a one-time US$57 in April 2021 to US$1,599 for a lifetime plan in 2026.
Read the storyThe founder had 10,000 followers on X who watched him build it, so it had 100 paying users two days after a 12-hour hackathon and sold itself by word of mouth after that. A no-code stack kept costs near zero, a lifetime deal whose price rose from US$19 to US$150 funded the first year, and yearly-only pricing cut support work for one person.
Read the storyThe founder sold places from design mockups to people in her own network before the product existed, and her Forbes column brought a quarter of the sign-ups at no cost. It sold creators what a general chatbot did not give them: the contact details of the people who chat, the transcripts, and the clone on their own site with their own brand.
Read the storyIt started two months after OpenAI's Whisper API made accurate transcription cheap, and it sold the step after the transcript: a summary, minutes, an email or a post, sent on to Notion, Zapier or WhatsApp. In 2023 it charged by the year, US$99 or US$199.
Read the storyIt solves one narrow problem for a very large app: WhatsApp users who chat with people in another language and do not want to copy every message into a translator. The app has been on Google Play since June 2021 and counts more than 700,000 installs. The Flippa listing says all growth was organic, that the translation API costs about US$20 a month, and that the app needs about two hours of work a week.
Read the storyThe founder built it for a problem he had himself as a new guitarist, so he knew the problem without market research. He shipped a simple web version first and charged from the start. Then he posted short videos of the product several times a day, with his own face, and one of the first videos spread widely. The iPhone app came only after the web version had paying users.
Read the storyOpenAI published its Whisper speech model as open weights, which anyone may run for free. Jordi Bruin, a Dutch developer of many small Apple apps, put it in a simple Mac app with drag and drop, gave the basic version away on Gumroad and charged once for the larger models. A newsletter, startups.email, reported in August 2023 that the app had made about US$100,000 in total, but it gave no source for that figure.
Read the storyThe company says Apple's built-in dictation "is prone to accuracy errors", and that typical dictation and note-taking apps keep recordings on their own servers. Voibe offered on-device dictation tuned for people who write long prompts to AI tools. The founder posted that most of the money came from lifetime licences, and that he planned to raise prices and move revenue to subscriptions.
Read the storyIt came out on 11 January 2023, when ChatGPT "has been all over the headlines recently", as MusicRadar wrote that day, and music writers were curious about any new use of OpenAI's models, and articles in MusicRadar, Fast Company and other outlets followed. The founder said in April 2023 that most users came from his posts on Twitter and that the press coverage kept bringing new people. He also said it took two weeks after the January launch to reach US$2,000 in revenue.
Read the storyIt did one simple job that many people wanted: get the content of a long video without watching it. The founder built it for himself as a weekend project, posted it on Reddit, and, as he wrote, sold the first subscription three days after launch. The current owners, who bought it later, describe it as a tool that did one thing well.
Read the storyThe next verified one-person AI business goes here. Same structure, same evidence grades.
Browse by what the product does
Patterns to copy
Every number on these cards carries an evidence grade. How Aiolum verifies explains the four grades and the five parts of each story.