Compare the signals
| Signal | What you can use it for | What remains uncertain |
|---|---|---|
| Your recording | Compare two attempts at the same phrase. | You may not yet hear the feature that needs work. |
| Recognised words | Check what text a recognizer detected. | Detection does not diagnose vowel quality or accent. |
| Sound analysis | Inspect the model’s estimate of a sound feature. | Accuracy depends on the system and recording conditions. |
| A teacher’s response | Get an explanation and a practice priority. | Ask which feature the judgment concerns. |
The tool comparison explains how different practice formats use these signals.
Turn feedback into a practice routine
Choose one short line from the phrase library. Record it once, then identify a specific change: one vowel, one consonant or one phrase boundary. Record the same line again. Comparing different sentences adds too many moving parts.
When a response is vague, ask for a smaller target. “Speak more naturally” leaves you guessing; “keep this vowel steady” gives you an action to test. Listen to a reference in the sound chart when you need an isolated model.
Common mistakes when reading feedback
Do not turn one missed word into a verdict on your French. Background noise or recognition errors may explain it. Likewise, all words being recognised does not prove that every sound matched the reference.
Avoid correcting several features because they appeared in the same report. Choose the one that prevents your intended meaning from getting across, then check it in another phrase.
Feedback inside Parle
Parle provides model audio, recording, replay and Phrase Match word detection. Phrase Match is not an accent, fluency or CEFR score. Optional AI Coach replies use text or speech transcripts, rather than analysing the recording for sound quality. Generated replies can contain mistakes.