Silence Detector & Remover

The detector slides a 20-millisecond window across the first channel of the decoded audio (window length is computed as sampleRate × 0.02) and computes the RMS energy of each window rather than a single peak sample, which keeps it from being fooled by isolated clicks or DC spikes. A window counts as silent when its RMS falls below a linear amplitude derived from your dB threshold via 10^(thresholdDb/20) - so the default -40 dB becomes roughly 0.01 amplitude, -50 dB about 0.0032, and -30 dB about 0.032. The threshold slider spans -60 to -10 dB in 1 dB steps, with three presets wired to fixed values: 'Very Sensitive' (-50 dB), 'Normal' (-40 dB), and 'Less Sensitive' (-30 dB, only true silence). The yellow band drawn on the 800×150 waveform canvas shows exactly where that threshold sits, and red overlays mark every detected silent region.

A run of silent windows is only recorded as a silence segment if its total length is at least the Minimum Duration you set (slider 0.1 s to 5.0 s, default 0.5 s). This minimum is the safeguard against false positives: natural inter-word gaps in speech are typically 100-200 ms, so a 0.5 s floor ignores them while still catching deliberate pauses. The end-of-file case is handled explicitly - silence that extends to the last sample is closed out and counted if it meets the minimum - so trailing dead air is detected, not dropped. The component reports total duration, count of silences found, total silence time, and a silence-ratio percentage, plus a scrollable table of every segment with start, end, and duration timestamps and a per-row play button that seeks to 0.5 s before the gap.

Two downstream actions operate on the detected segments and both encode to MP3 via lamejs 1.2.1 at 192 kbps. 'Remove All Silences' rebuilds a single AudioBuffer containing only the non-silent stretches (preserving the original sample rate and channel count) and downloads one trimmed file. 'Export Audio Segments' instead slices each non-silent region into its own MP3 - useful for splitting a recording into tracks or clips at the silence boundaries - using a 0.01 s guard so razor-thin fragments aren't emitted, and inserting a 300 ms pause between each browser download. Everything runs in the browser through the Web Audio API; supported inputs include MP3, WAV, OGG, AAC, M4A, FLAC, WebA, MP4, and WebM. Note that breathy or reverberant recordings carry a high noise floor, so a strict threshold like -40 dB may never register their pauses - raise it toward -30 dB for such material, or lower it toward -50 dB to catch quieter gaps.