IELTS Speaking Coach
Local-first IELTS speaking practice, end to end
IELTS Speaking Coach is a practice loop, not a chatbot. You hear the question, you speak, it transcribes what you actually said — then tells you what to fix, with the evidence attached. Everything runs on your own machine.
How a session runs
Listen
The question is read aloud. Part 2 comes with a 60-second prep timer and a scratch pad.
Speak
Keyword prompts act as scaffolding — enough to get you talking, not enough to read off.
Transcribe
What you said becomes text you can look at — the thing that normally vanishes the moment you stop talking.
Review
Grammar diffs against your own sentences, pronunciation diagnosis, and side-by-side comparison with your earlier takes.
Depth on one question, not chat with an examiner
Most tools put an AI examiner in front of you and let you talk. This one does the opposite: the same question, recorded again and again, each take compared against the last so progress is visible. Keyword scaffolding trains the real exam skill — speaking from points — instead of memorising a script.
Scores that have to show their evidence
A score you can argue with is worth more than one you just have to trust. A full 14-minute mock ends in four-criteria band estimates, and every conclusion has to quote your transcript or an objective metric — so you can see what it reacted to and judge whether it got you right. Pronunciation is deliberately not handed to the model: the model only sees text, and text contains no pronunciation. That one is computed from acoustic evidence instead, with a confidence level printed next to it. It is an estimate, not an official score.
Drills built from your own mistakes
You drill the words you actually said and actually got wrong — a better use of an evening than working down a generic minimal-pair list that has no idea which sounds give you trouble. Anything in your transcript that lands on a weak phoneme goes straight into the queue.
Nothing leaves your machine
Practise as much as you want without watching a meter, and without handing your voice to anyone. There is no account to create and no usage to pay for: speech recognition runs on your own GPU, and language models go through CLI subscriptions you already have. Your recordings stay on your disk, which also means it keeps working on a plane.
What's inside
Library
Topics, questions, keywords. Part 2 cue cards are modelled separately.
Practice
Listen, answer, transcribe, get grammar feedback graded against Band 6.
Mock exam
Recording starts on its own, pauses are picked up, no retakes — one report at the end.
Pronunciation
Minimal-pair diagnosis, ordered by how much each one hurts comprehension.
History
Same question over time, a grammar mistake book, and trend lines.
Download
Packaged desktop builds are on the way. Speech recognition runs on Apple GPUs, so macOS comes first.