Naramine: Giving Web Novels a Voice Without Leaving the Page

My CV does not tell one story, it tells a dozen short ones. I moved between Java, C#, and Ruby on Rails. I spent four years as a teaching assistant at the Politehnica in Timisoara, started a PhD I never finished, taught kids to code for three years in a collaboration with Nokia, and ran an outsourcing firm through the peak crypto years with clients and operations spread across the world. Somewhere in there I wrote a fantasy novel and did a stretch of investing in the stock market. For a long time I read that list as a lack of discipline. Eventually I read it correctly: I am a starter. I love taking things from zero to the point where they produce results, and I get restless in the long steady middle. Naramine is the first product I have designed around that fact instead of against it: a browser extension that reads web novels aloud, on the page where they live.

A starter, on purpose

The realization about being a starter did not arrive as a business insight, it arrived as self-acceptance. Once I stopped treating every unfinished decade-long ambition as a failure, the pattern looked less like flakiness and more like a preference with rules: I am at my best initiating something, pushing it until it works, and handing the marathon to someone or something else. That reframe produced a list of things I wanted to have done, the fantasy book, the investing period, and, at the top, a product of my own. Naramine is that checkbox, but it is also a test of whether I can build a company shaped like me.

Half a billion views and no play button

The stories I wanted to consume are not on Audible. They are on Wattpad, Royal Road, Archive of Our Own, and Honeyfeed, self-publishing platforms that together pull roughly 500 million views a month and serve as the launch ramp for authors the big publishing houses ignore. I did not want to read them on a screen, I wanted to listen, while walking, driving, doing dishes. The tooling for that is surprisingly bad.

Generic text-to-speech products exist, Eleven Labs and Speechify among them, but they share a structural flaw: they pull you out of the site. You extract or import the text and consume it in their app, which makes them traffic devourers from the platform’s point of view, an invitation to litigation rather than partnership. And they do not understand the shape of a web novel: chapters, pagination, the need to autoplay into the next installment at two in the morning. The voices, where free, tend toward the soporific. So for a long time Naramine existed as a private backend, an engine only I used, feeding my own listening habit. In the last three months I gave it an interface and turned it into a product ready for beta.

Chunks, buffers, and the next chapter

The demo I show people is mundane on purpose: open a story, press play, and the chapter starts speaking within seconds. Getting to mundane took engineering. Audio is generated chapter by chapter, and the first chunk of each chapter is deliberately tiny so playback can begin almost immediately; the chunks then grow until the pipeline settles into normal flow. While you listen, the system pre-buffers ahead, and when a chapter ends it has already begun preprocessing the next one in the background, so autoplay between chapters feels like one continuous audiobook rather than a series of loading screens.

Everything generated lands in IndexedDB, in the listener’s own browser. A heavy user’s cache can reach tens or even hundreds of megabytes, so there is a clear data control and a download option, but none of it ever touches my servers.

The same engine twice: Kokoro as free and as paid

The voice engine is Kokoro, open source, and I run it in two very different ways. Basic mode is Kokoro JS executing locally in the listener’s browser: on my Mac it synthesizes at about 1.3 times listening speed, which is enough to stay ahead of the narration, but on weak hardware it can drop below 1.0, around 0.8 on some machines, and the audio stutters while the pipeline catches up. It is also limited to English. Premium mode runs the heavier models on GPUs through Modal: cold starts of two to three seconds, audio delivered in chunks with the first one arriving very fast, better voices, additional languages, and no dependence on the listener’s laptop.

The economics are almost embarrassing. Processing 50 chapters on Modal costs under a dollar at my current setup, and Modal gives every account 30 dollars of free credits a month; I have noticed that multiple organizations under one account each seem to receive their own 30 dollars, a loophole I am treating as beta-phase budget rather than a business model. Sentry watches for errors. Heatmaps can wait.

Sparkles, and a ten dollar gift

Premium is priced in credits I call Sparkles: a typical chapter of around 10,000 characters costs about 20 of them. What I refuse to do is charge a monthly subscription. Web novel reading happens in waves, a two week binge on a story, then nothing for a month, then another binge, and a subscription punishes exactly that behavior, forcing people to cancel and resubscribe or quietly pay for silence. So the model is a one-time purchase, around 10 dollars, framed as a gift you give yourself, with the Sparkles lasting as long as they last. My cost behind a 10 dollar purchase is on the order of a dollar and a half, which keeps the margin near 80 percent, and I am deliberately staying away from the expensive commercial TTS engines that would eat that margin alive.

A fair, not a funnel

My marketing plan starts with a Discord, and the Discord starts with characters. Before scale, before ads, I want the server to feel like a fair, several stalls, each run by a personality: a master of ceremonies who makes announcements, drops tips, proposes events. In the beginning those characters will be me, performing on several voices, which given what I am building feels appropriate. Later, friends can take over roles, or a part-time community manager. Members get a one-time gift of Sparkles for joining, financed by those Modal free credits, and the calendar gets challenges, events, and loyalty mechanics designed to grow a nucleus of users who evangelize the product because they feel like citizens of it.

Only then comes promotion, in a deliberate order: Reddit first, organically, building reputation by being useful in the web novel communities rather than shame-plugging the product; the Discord catches and keeps whoever arrives. Paid ads are a maybe, an experiment for later, possibly targeting keywords and genres, and I have already talked myself out of using the platforms’ logos in ad creatives, which implies partnerships that do not exist yet.

Everyone googles “listening”

The SEO plan is unglamorous and specific. People do not search for “text to speech for web novels”, they search for “Wattpad listening” and its siblings, platform by platform. Those queries are long-tail, intent-rich, and much less contested than the generic TTS keywords the big players fight over. The plan is keyword research first, including a hard look at what competitors like Eleven Labs already rank for, then articles targeted at the listening intent for each platform I support, published at a cadence closer to one a day than one a month, starting before launch so the compounding starts early.

The platforms are the endgame

Everything above is shaped by one strategic choice: Naramine stays an accessibility tool, and I want relationships with the platforms rather than a war. That rules out mass scraping and the kind of features that antagonize the people whose content my users love. It also dictates the partnership sequence. Wattpad belongs to a very large East Asian corporation, and an unlaunched solo founder does not get that meeting. The smaller platforms, still counting their users in the millions, have the same reader-retention problem and answer their own email. The plan is to win one of them as a proof of concept, an official integration, engagement numbers, authors who like what it does for their audience, and then take those receipts into the bigger conversations. The stakeholder list is already built on LinkedIn, and the outreach materials are drafted.

Mobile is the elephant

I will say the uncomfortable number out loud: 80 to 90 percent of the people reading on these platforms are on their phones, and Naramine today is a desktop browser extension. Chrome first, then Edge and the other Chromium browsers. Firefox does allow extensions on mobile, but its curated catalog is limiting; Safari on desktop is plausible, Safari on mobile is uncertain. A native mobile app is clearly a necessity, and it is deliberately later work, because the hard problems there are real: autoplay and chapter navigation with a locked screen, inside ecosystems built to prevent exactly that kind of background behavior. I would rather nail the desktop experience, prove the model, and arrive on mobile with revenue than fight Apple and Google on day one.

Two weeks out

Naramine leaves beta in about two weeks, two sprints from now. The sprints have names: gamification and polishing. The polish list is honest, one essential button still needs the right name, tooltips and microcopy need full sentences, and the extension’s install-time permissions are being audited down to the minimum, because nothing kills an install like a scary permissions dialog. Sentry is already watching for errors. Then it ships, and the starter in me gets what he came for: real users, real feedback, and a product that produces results while I figure out which parts of the marathon to automate, delegate, or, knowing myself, hand to a character with a better voice than mine.