A technical breakdown of client-side Web Audio decoding, RMS loudness analysis, non-destructive pause mapping, and zero-upload audio rendering.
When you select a file, AutoSilenceCut reads it locally via FileReader into memory. The browser's native C++ media engine decompresses the audio into planar 32-bit floating-point PCM buffers with zero network uploads.
Rather than relying on naive peak thresholds that clip quiet consonants, AutoSilenceCut calculates Root Mean Square (RMS) energy across 20ms sliding windows, continuously tracking ambient room tone.
RMS = sqrt( (1 / N) * sum( sample[i]^2 ) )Traditional silence strippers delete 100% of pauses, making dialogue breathless. In Natural mode, AutoSilenceCut retains a calibrated 0.32s cushion around spoken phrases to preserve authentic human pacing.
Abrupt waveform cuts create harsh DC offset clicks. AutoSilenceCut automatically applies 8ms equal-power linear crossfades at every edit boundary, guaranteeing imperceptible, click-free audio splices.
Every detected gap is inspectable. Listen to before/after previews and restore individual pauses instantly with full 50-step undo/redo support before exporting.
OfflineAudioContext renders spliced audio directly into uncompressed 16-bit PCM WAV format at native sample rates, avoiding destructive MP3 re-encoding.
Try the browser workspace with your own recording or test with our sample audio.