harlan-zhang/maxspeak

MaxSpeak is a free, open-source web application that turns text into spoken audio using a third-party AI voice service called MiniMax. You type or paste text, choose a voice, and the app generates a spoken recording you can play back or download. It runs in a browser and requires you to supply your own MiniMax API key, which you paste into the settings screen when you first open it. The app covers three main tasks. First, text-to-speech: you write something and pick from more than 300 preset voices across 20-plus languages, including Chinese, English, Korean, Japanese, Spanish, and several European languages. You can adjust the speed, pitch, and tone, choose one of nine emotional styles such as happy, sad, or angry, and add short sound cues like sighs, laughter, or breathing by typing simple tags inside the text. Second, voice cloning: you upload about ten seconds of someone speaking, and the service produces a new voice profile that mimics that person. Third, voice design: you describe the voice you want in plain words, such as "warm and gentle, suitable for a bedtime story," and the AI generates a matching voice from that description. Long pieces of text stream as audio in real time, so playback starts before the full file is ready. You can also wait for the complete file and download it in formats like MP3, WAV, or FLAC at various quality levels. The app keeps your API key on the server side so it is never exposed to the browser directly. The project is built with Next.js and TypeScript and can be deployed to Vercel in one click. Cloudflare Pages is also mentioned as an option, though the README notes that some features do not work there. Everything runs through Next.js API routes that sit between your browser and the MiniMax servers. MiniMax charges per use on a pay-as-you-go basis, costs for text-to-speech are listed in the README in Chinese yuan. MaxSpeak itself is MIT licensed and has no subscription of its own.