Count your comprehensible input hours.

Serval is a free Chrome extension that tracks comprehensible input: the hours you spend listening to a language you're learning, counted the way immersion roadmaps mean them. It runs on YouTube, Twitch, Kick, Vimeo, Dailymotion, TED, Bilibili, Niconico and Netflix, reads time from the video player's own clock so that ads, pauses and seeks are never counted and a hidden tab still is, and listens to each video to identify its language, one of 105 it can recognise, entirely on your computer. Video in a language you already speak is recorded but held out of your total, with the held-back time shown beside the number it came out of, so nothing silently disappears. When a video's language can't be identified, its time isn't counted; nothing is guessed. Everything is stored in your browser, with no account, and can be exported, backed up and restored. It runs in Google Chrome, Brave, Edge and other Chromium browsers, and in Firefox, on a computer.

Because listening improves too slowly to notice from one week to the next, and that is when most people give up. A total that keeps rising shows the work is adding up when it doesn't feel that way, and hour landmarks tell you what to expect next. Tracking doesn't teach you anything by itself, in the same way a step counter doesn't walk for you, but seeing the number makes it easier to keep going. Serval does the counting so there is no log to remember.

Anyone learning a language by watching and listening, whether that is your main method or something you do alongside classes and apps. It suits people following comprehensible input or immersion approaches, such as Dreaming Spanish or Refold, and anyone who wants to know how much of a language they really hear. If you only study with textbooks or flashcards, it has little to count.

Not by playing it overnight. There's no good evidence that audio during sleep builds real understanding. Sleep does help in another way: studies suggest that the new words and sounds you heard during the day settle in overnight. So listen while you're awake, then get a good night's sleep. Serval can't tell if you've dozed off, so a video left playing still counts; a sleep timer keeps your total honest. There is more in what sleep really does for language learning.

Yes, and there is no catch to find. There is no account to create, no subscription, no paid tier and no advertising. Everything the extension records stays in your browser, so there is no data to sell and no server bill to pass on. The counting, the language detection, the streaks, the calendar, the achievements, the backup and the export are all included for every user. If Serval ever has a paid tier it will be for something new, such as a sense of how hard each video was, and never for the counting itself. Nothing you have today will be taken away or put behind a price.

Serval runs in Google Chrome and in the other browsers built on Chromium: Brave, Microsoft Edge, Vivaldi, Opera and Arc; and in Firefox, version 140 or newer, on a computer. Time is counted and languages are identified in all of them. The Firefox version is the same extension, installed from Firefox Add-ons, and opens in Firefox's sidebar rather than a side panel. In every one of them the analysis runs entirely on your computer, so no audio leaves your machine; the privacy policy has the detail. Safari uses a different extension system and isn't supported, and phones aren't either, because a browser extension cannot see what a phone plays. If you use one of the supported browsers and a label never appears on a video, the support page lists what to check and how to reach us.

Nowhere. Serval never records audio, never stores it and never sends it anywhere. To identify the language, short windows of the video's sound, up to thirty seconds at a time, are analysed on your own computer by a small model that ships inside the extension, then discarded; the only thing kept is the language label for that video. Nothing about what you watch is sent to us either: no video titles, no addresses, no languages, no durations. Your history lives in your browser and nowhere else. The full detail, including the handful of anonymous usage events the extension sends about itself, is in the privacy policy.

105. That includes Japanese, Korean and Mandarin; Spanish, Portuguese, French, Italian, German, Dutch and the Nordic languages; Russian, Ukrainian, Polish, Czech and the other Slavic languages; Arabic, Hebrew, Persian, Turkish, Hindi, Bengali, Tamil and Thai; Vietnamese, Indonesian, Malay and Tagalog; Swahili, Amharic and Yoruba; and many more. You never have to say which one you are learning: the language of each video is worked out by listening, so you can learn several at once and the panel breaks the day down by language. The one thing you declare is the languages you already speak, so their time is held out of the total. Twenty-eight of the 105 have been measured against real recordings, all correctly; the rest rest on the model's published results, and every label still needs two independent checks to agree before it is shown. The full list is at servalinput.com/languages.

Conservative by design. A wrong label that lands on your native language would remove hours from your total, so Serval would rather say nothing than guess. A label is only shown once two independent checks agree, and until then the time waits: it joins your total the moment the label lands, and never if it doesn't. Measured against recordings in 28 languages, every one was labelled correctly on thirty-second windows, and close pairs such as Italian and Spanish or Korean and Japanese were not confused. The remaining weakness is a voice buried under loud music, which may sit unlabelled for a while; on YouTube the video's own caption language settles those. If a label is ever wrong, the video's address is the most useful thing you can send to support.

A voice buried under loud music can't be identified from the audio alone, so those videos may go unlabelled, and unlabelled time is not counted. On YouTube, Serval falls back to the language of the video's auto-generated captions after a couple of minutes, which is YouTube's own reading of the clean audio and usually right, and the time counts from the start of the video once it does. On other sites a video that never gets a label never counts. Background music at a normal level, the kind under most talking-head videos and podcasts, is fine; it is specifically speech that is quieter than the music that defeats it.