A free English app for Turkish speakers. Your lecturer’s English is transcribed on your phone; you learn with songs, speaking practice and spaced review. Turkish stays beside you where you get stuck, and steps back as you grow.
Animation: an Operating Systems lecture in the Lecture Desk. English sentences stream in; the Turkish line shows according to the chosen level: under every sentence, only on tap, or off.
Beside. English first, Turkish under every sentence. Terms stay in English; their Turkish appears in brackets only the first time.
Sped-up animation with example sentences. In the app text arrives in chunks of about 30 seconds. The app’s interface is fully available in English.
Counted from the content of app version 0.8.1.
Why SpeakSmith
The transition zone
In ecology, the strip where two habitats blend is called an ecotone. SpeakSmith builds one from Turkish to English: you are always inside English, you choose how much Turkish support you use, and you pull it back as you get ready.
Turkish under every English sentence. Terms stay in English with their Turkish in brackets the first time only.
Turkish hidden. Tap “TR” on the sentence you are stuck on; only that one opens.
English only. Tapping a word still opens its card. The exam will ask for the English anyway.
In improv the situation is given in Turkish and you answer in English; hints open only when you ask. In presentations you speak for 1–3 minutes.
The lecture summary tells you, with a measured number, how much you relied on Turkish. After three lectures with very little Turkish, the app suggests moving one level up, once; the level never changes by itself. When you reread a lecture you pick Both, EN or TR first; even in TR first the English stays on screen.
One app
Every card is a feature that works in today’s version. Tags tell you what it needs.
Transcribes a live English lecture on the phone, offline. Keeps going with the screen locked.
A captioned lecture video becomes a lecture, sentence by sentence. In YouTube: Share › SpeakSmith › “Watch as a lecture”.
23 songs on shelves by level. Word-by-word fill and the line’s Turkish.
Loop a weak line: listen, sing along, then solo. Compare your voice with the original.
The song stops at the gap; pick or type the missing word. No microphone needed.
120 scenes. A Turkish situation, a spoken English answer. If the character doesn’t understand you, it says so.
Talk for 1–3 minutes; see your pace, filler words and presentation functions right away.
Listen, repeat, and see a word-by-word comparison.
FSRS-5 spaced repetition, 6 exercise types, 2,800-word level packs.
33 castles, 208 mini exercises, 99 tips; 33 of them are “Turkish traps”.
138 patterns across 15 functions such as suggesting or politely disagreeing.
112 steps across 8 tracks; each step has a short lesson, the next one waits on the home screen.
Lecture Desk · Android
The lecturer speaks; sentences appear on screen. The on-device Whisper model does the transcription: after a one-time download no internet is needed and audio never goes to a server.
Operating Systems (118 terms) and Differential Equations (68 terms) come as ready templates. Paste your syllabus or pick its PDF and term candidates are found offline; you choose which to add. A term with no known Turkish stays blank instead of being machine-translated.
In “Far (classroom)” mode the phone uses a suitable microphone source; audio is filtered and boosted up to 12×, and silent chunks are never sent for transcription. You pick the Turkish line level and the translation route here, and optionally keep the audio (90 minutes ≈ 21 MB).
Recording and transcription continue with the screen locked or in another app, with a quiet notification showing the time and a Pause button. Text is saved every 30-second chunk; if the app closes mid-lecture, the rest is transcribed on the next launch.
Back at the desk it tells you plainly what happened, how many new lines arrived, and if recording stopped it says so and offers “Continue”. Sentences you flagged with “Lost me” (placed 2 seconds back) land under “Questions for the lecturer” in the lecture notes, with or without AI. Later you can reread it as Both, EN or TR first, and have its Turkish lines translated again.
Good to know: transcription is not real-time; the first lines arrive after about 30 seconds. The models transcribe English speech only. Ask your lecturer and school before recording.
Karaoke · a 23-song catalogue
23 songs from A1 to C1 on shelves by level, yours first. One tap prepares the lyrics and opens the player. Words fill as they are sung, with the line’s Turkish underneath if you want it.
We walk along the quiet river
The city lights are on tonight
And every word comes back to me
nehir
Press play, then tap a word. In the app you get its meaning in context and one tap adds it to your list.
A song’s level is estimated from its words, repetition and tempo, and is labelled “estimate”.
Tap a line in the result or “Practise the 3 weakest lines”: the line loops, and tempo rises on good rounds.
Long-press a line: your voice first, then the original. The result stays in the player.
After the song, the words you sang are compared with the lyrics: a percentage per line, A–F for the song. A chorus you skipped doesn’t score full marks. Without headphones the original vocals can leak into the mic; that score is marked “estimate” and earns no bonus.
And every comes back to me
The song stopped here. Which word fills the gap?
Good to know: the catalogue holds only title, artist, level and tempo; lyrics and audio are not bundled. Songs play in the in-app YouTube player (internet needed), with lyrics from YouTube’s captions or LRCLIB. On Android, scoring and line practice use a one-time Whisper download and your recording never leaves the phone. In the browser the browser’s speech recognition scores you; unless you confirm “My headphones are on”, the score counts as an estimate with no bonus and no best score.
Speaking
If the character really doesn’t understand you, it says so. Example chat; needs an AI coach.
Without AI, right on the phone: words per minute, which filler word how often, and how many of 6 presentation functions you built. Topic chips need no AI either.
Example numbers. If speech recognition or the mic permission is unavailable, you carry on by typing.
It checks the words you said; it is not a pronunciation or phoneme analysis.
In shadowing, presentations, improv and spoken word review the mic turns off as soon as the phone locks or a window opens on top, and it doesn’t turn itself back on when you return.
Words · Grammar · Patterns
FSRS-5 schedules each word’s next review. Due reviews always come first; new words join at your daily limit (10 by default). Words from lectures, songs and chats meet in the same deck.
Each topic is a castle. Read the lesson, pass the offline mini exercise and the castle falls. When review is due it is marked “needs upkeep”.
Bars show an example progress.
A little every day
The home screen tells you what to do today: the next path step, your daily goal, today’s plan. Day one starts with a small win that needs no AI. One missed day a week is bridged automatically, and a streak at zero simply says “Start your streak”; after a break of three days or more you get a welcome with no blame and +30 XP. Three small quests a week, doable without an AI coach, and 27 badges in 8 categories. Importing a word pack or CSV doesn’t fill your daily goal: a word earns its XP on its first successful review.
Privacy and data
Progress, words and lectures stay on your phone and are excluded from Android cloud backup. Some features connect to third-party services (YouTube, LRCLIB, dictionary and translation services, model downloads and, only with your own key, cloud AI). The speaking practices use the device’s speech recognition service, which may send audio to its provider when no offline English pack is installed, and read-aloud prefers natural voices, which are often online voices from Google or Microsoft.
In the browser version your data is stored in this browser under SpeakSmith’s own keys, and “Delete all data” removes only SpeakSmith’s data. Clearing site data for mehmetyasinuzun.github.io in your browser settings also clears the developer’s other apps on that address (such as SevimliPati). Web API keys are encrypted, but the key that decrypts them sits in the same browser storage, so don’t enter a cloud key on a shared computer.
Download
Version 0.8.1 · October 2026 · signed APK. When a new version is out the app tells you; the same signature lets it install on top and keeps your data.
Point the camera at the code: this download section opens on your phone, and its button downloads the APK straight to it.
This is an iPhone or iPad: the APK cannot be installed here. Open the browser version, or open this page on an Android phone.
Compare the SHA-256 digest of the file you downloaded with the one on GitHub’s release page, which lists it next to every file.
speaksmith-arm64.apk, the file the button downloads (the same file as speaksmith-v0.8.1-arm64.apk on the release page):
8fcaa1dbb9334595ad63745a0d2037db6533062cc88bd8e2d422af6f77ee0b9d
Won’t install? Versions before 0.5.0 were signed with another key: back up in Settings › Data management, uninstall the old app, install the new one and restore the backup on first launch. The universal APK is only for a first install on an unusual device: it does not install over an existing arm64 or arm32 SpeakSmith (Android treats it as an older version).
Coming from 0.7.0 or older? Those versions may not show update notices. Install the current version on top once from here; after that the app tells you about new versions.
Word review, the grammar map, patterns and the learning path work in the browser. Your data is stored in this browser, with no account or sync. Depending on the browser, the first load downloads about 7–14 MB; prefer Wi-Fi on mobile data.
Open in the browserFAQ
Yes. No fees, no ads, no in-app purchases. An on-device AI coach is free too; if you connect a cloud provider, that provider bills your own account.
Once the speech model and the Turkish translation pack are downloaded, live lectures are transcribed and translated offline (unless cloud AI is selected). Word review, grammar, patterns, the path and the on-device AI coach also work offline. Adding YouTube content, playing catalogue songs, dictionary look-ups and cloud AI need internet.
No, and we don’t claim to. Shadowing and karaoke transcribe what you said and compare it with the expected text, so they check whether you said the right words. There is no waveform or phoneme-level pronunciation analysis.
No. The on-device speech models transcribe English speech only. Lecture Desk is built for lectures taught in English.
The catalogue holds only title, artist, level and tempo. Lyrics and audio are not bundled: lyrics come at play time from YouTube’s English captions or LRCLIB, and the video plays in the in-app YouTube player. The option to also try downloading YouTube audio for offline use starts switched off, because YouTube usually doesn’t allow it.
You can add your own audio file too. If you pick “AI transcript (Whisper)” as the lyrics source while adding it, the on-device Whisper model writes the lyrics on Android; with no model on the phone (always on the web), the file is sent to OpenAI Whisper if your OpenAI key is saved. On a song you already added, “Get lyrics from the audio” uses only the on-device Whisper model (Android).
Yes. Every screen is available in Turkish and English; you choose in onboarding or in Settings. The app is designed for Turkish speakers, so Turkish support (translations, hints) is built in and can be reduced.
On launch the app checks a public version file; if there is a new version it asks you and downloads the APK from the GitHub release page. No identity or usage data is sent.
If you use 0.7.0 or older, you may not see a notice: install the current version on top once from the download section, and the app will tell you about later versions.
Not yet. You can open the browser version (first load ~7–14 MB); speech recognition there works best in Chrome and Edge and is limited in Safari. Lecture Desk and Whisper features are Android-only.
Start in minutes, without an account. Stay inside English; never alone.