Every take,
scored to the phoneme.
The hardest part of speaking practice is having no one to talk to. SpeaKoach is a partner that's always up for it — it chats back in character, hands you three more natural ways to say what you just said, and scores your pronunciation down to the phoneme.
See the take the moment you finish speaking.
Transcript, score, and correction land together — weak words underlined inline, IPA one hover away, the whole line rewritten if you code-switched into Chinese.
Weak sounds light up. A shaky vowel gets an amber underline, a miss gets red — hover any word for the IPA you actually produced.
Three ways to say it. After every sentence: casual, native, formal. For when you know the words but the phrasing won't come out.
Stuck? Switch to Chinese. SpeaKoach understands the code-switch and rewrites the line at three registers.
Practice like it's a recording session, not a chat log.
Someone to talk to, anytime
Speaking more is the only way to get better — but most of us have nobody to practise with. Fifteen role-play scenes, free chat, and timed IELTS/TOEFL tasks, always in character.
Three ways to say it
You know every word, yet the sentence won't come out right. After each utterance you get casual, native, and formal rewrites — steal the phrasing, then drill it.
Weak sounds light up
Good sounds stay quiet. A shaky vowel gets an amber underline, a miss gets red — hover any word for the IPA and one concrete tip.
Code-switch, mid-sentence
Say it in Chinese when you get stuck. SpeaKoach rewrites the full sentence in clean English at three registers.
Takes you can interrogate
Every attempt is a numbered take you can rewind and redo. The progress page aggregates your weakest sounds, speech rate, and pauses — measured, not self-reported.
Local-first, zero cost
ASR, phoneme scoring, and TTS run on your machine. Bring any LLM — even a free tier — and your voice never has to leave home.
Up and running in three lines.
Docker is the fastest path. Open localhost:8000 once it's up — models cache on first run, and your sessions survive upgrades.
Prefer source? make install && make dev — UI on :5173, API on :8000. See the README for details.
"Say /ˈlɑːteɪ/ — two clear syllables, stress on the first."— the kind of tip SpeaKoach gives on your worst sound in a take