One word. The same example.

Listen. Repeat. Compare.

This separate experiment compares your recording directly with two saved examples. It does not use HuBERT or give a pronunciation grade.

The word to practise

last

Listen to the synthetic British example, wait for 3, 2, 1, then say last once. Your microphone stays paused during the example and countdown.

Ready when you are. No model download needed.

Microphone level Microphone off

Hear the two references

“Lost” is a comparison sample to check whether this method distinguishes the vowel. The voices are deliberately fixed for this experiment.

First try “last” naturally. Then make a separate attempt saying “lost”. No failed attempts are counted and nothing is added to a practice list.

Or compare a saved WAV (one word, up to 8 seconds):

What the result means

“Closer to last” means this acoustic comparison favours the example. It does not prove correct pronunciation. A different voice, microphone or room can affect the result. “Uncertain” is an acceptable outcome of this test.

Audio stays in this page’s memory and is discarded when you close it, unless you save it. Reference clips are bundled with this page. No recordings are uploaded.

Test details

MFCC sound features are aligned in time. Smaller distance means closer; these are not confidence percentages. The provisional decision needs distance to “last” below 0.20 and at least a 15% relative advantage over “lost”. There is no human calibration yet.