Medical English Pronunciation Drill

Select a preset word set, hit Record, and speak clearly. The tool compares your speech to the target using your browser's built-in speech recognition and highlights every word right or wrong — no server, no upload, completely private.

All sets are bundled — no network request needed.

How it works

Every word in the target phrase is normalized to lowercase and stripped of punctuation. The tool does the same to what the speech recognizer heard, then aligns the two sequences word by word. A word is marked correct if the recognized token matches exactly; otherwise it is marked incorrect.

Matching logic normalize(expected) === normalize(heard): lowercase, strip punctuation, trim whitespace.
Accuracy score correct tokens ÷ total expected tokens × 100%. Multi-word phrases count per token.
Weak word detection Any item answered incorrectly ≥ 1 time is added to the weak list and promoted in "Weak words first" mode.
Privacy Audio is processed by your browser's built-in recognizer (on-device or OS-level — never uploaded to Knackpad). Score history is stored in localStorage only.

Frequently asked questions

Which browsers support the microphone recognition feature?
Google Chrome (desktop) and Microsoft Edge have the most reliable Web Speech API implementation. Safari on macOS 14 Sonoma and iOS 17+ also works. Firefox does not currently support SpeechRecognition. If you see a yellow warning banner at the top of the page, your browser is unsupported — switch to Chrome or Edge for the full experience.
Is my voice audio sent to a server?
On Chrome and Edge the SpeechRecognition API sends audio to Google's or Microsoft's speech service respectively — this is handled entirely by the browser, not by Knackpad. Knackpad itself never receives your audio or transcription. No data is stored anywhere outside your own browser (localStorage for score history). If you need fully offline recognition, the tool falls back to manual text entry when the API is unavailable, and an offline Whisper-tiny ONNX model can be loaded optionally (see the Import CSV tab note).
How do I import my own vocabulary list?
Click the "Import CSV" tab, then either upload a .csv or .txt file, or paste directly into the text box. Each line should be one item. You can optionally add a second column (comma-separated) with a phonetic hint that will be displayed under the target word during the drill — for example: tachycardia, tak-ih-KAR-dee-uh. A third column can hold a sample sentence for context.
What does "Weak words first" mode do?
The tool tracks every item you have answered incorrectly across sessions (stored in localStorage). In "Weak words first" mode it puts those items at the front of each new session so you drill your problem areas more often. You can view the current weak word list at the end of any session, and clear your history at any time with the "Clear History" button.
Why does the recognizer sometimes mishear a medical term?
Medical terminology (e.g., "dyspnoea", "haemoptysis", "myocardial") is phonetically unusual and general-purpose speech recognizers sometimes substitute common words. Speak slowly and enunciate clearly — this is itself good pronunciation practice. The phonetic hint shown under the target word shows the stressed syllable in caps to guide you. You can also try the American English spelling variant (e.g., "dyspnea") as recognizers trained on US English handle those better.