A PRACTICAL NOTE · Troubleshooting

How to apply this guide

Recognize clipping by replaying a short sample, then reduce input gain or adjust distance.

Start with

Record the same sentence at a natural level and a slightly greater distance. Look for consistent distortion on louder syllables on playback. If an editor shows a flattened waveform, lower gain at the first adjustable input stage.

Check before you finish

For a source file that is already clipped, trim silence if needed and check the original against the transcript. Do not use rewriting to guess words that were lost in the audio.

Evidence to keep: Record the expected sentence, actual result, microphone, language, model and app build before changing settings. This page’s key limit is: Recognize clipping by replaying a short sample, then reduce input gain or adjust distance.

Start here

Clipping can make speech sound harsh or fuzzy even when the meter looks strong. A larger speech model cannot fully restore a syllable flattened at capture. Use a phrase that includes both soft and emphatic words to tell the difference.

Make a clean reference

Record the same sentence at a natural level and a slightly greater distance. Look for consistent distortion on louder syllables on playback. If an editor shows a flattened waveform, lower gain at the first adjustable input stage.

Keep small changes comparable

Change either the mic position or the gain, test once, and write down which adjustment helped. Check names and the last word. A loud waveform alone does not confirm good speech recognition.

Avoid post-capture guesses

For a source file that is already clipped, trim silence if needed and check the original against the transcript. Do not use rewriting to guess words that were lost in the audio.

Keep exploring