Detecting 20 ms silence windows with 80 ms speech padding
Breaking podcasts, voice memos, or live recordings offline without uploads keeps private interviews off remote servers. Choosing By silence scans quiet gaps between spoken phrases automatically so long recordings divide into clean tracks without manual timestamp scrubbing.
Decoding MP3, WAV, and M4A tracks locally
Clicking Load sample multi-part audio generates a 5.5-second speech clip (sample-multipart.wav) containing three spoken phrases separated by 0.9-second silent gaps. You can also drop WAV, MP3, M4A, OGG, FLAC, WebM, or AAC files between 4,000 Hz and 192,000 Hz onto Drop an audio track to split.
Uncompressed 8-bit, 16-bit, 24-bit, and 32-bit RIFF files parse in JavaScript, while compressed formats decode via the MDN BaseAudioContext decodeAudioData API. If you want to tighten speech pauses without splitting tracks into separate files, use the silence remover instead.
Tuning Silence threshold (dB) and Minimum pause (ms)
Inside splitAudioBuffer, the scanner steps through non-overlapping 20 ms frames and records the maximum absolute sample amplitude across all channels. Any contiguous run of frames quieter than Silence threshold (dB) (-60 dB to -15 dB, default -38 dB) forms a candidate cut zone.
Internal split boundaries add 80 ms of speech padding after a phrase finishes and 80 ms before the next phrase begins. Only padded gaps lasting at least Minimum pause (ms) (150 ms to 3,000 ms, default 400 ms) split the buffer, while tracks with zero qualifying pauses safely return Part 1.
How do chapter timestamps and 10 ms edge fades prevent clicks?
Chapter lists divide continuous recordings using explicit MM:SS or HH:MM:SS cue markers even when background music fills every pause, while 10 ms raised-cosine fades pull slice edges down to zero amplitude so standalone clips play without audible pops.
Parsing YouTube and podcast chapter timestamps
Selecting Chapter timestamps displays Chapter timestamps (one per line) alongside an Insert example chapters helper. Running inside everyday.worker.ts via the MDN Web Workers API documentation, parseChapterTimestamps accepts leading cues (01:15 Topic), bracketed timestamps, decimal milliseconds, separator characters (-, :, |), and trailing titles.
Lines starting with # and blank entries are skipped, markers sort ascending, and cues within 50 ms deduplicate cleanly while intervals under 20 ms are discarded. When building podcast episodes before slicing chapters, you can combine voice segments with music beds into one master mix beforehand.
Using Fixed interval slicing and 10 ms cosine fades
Switching to Fixed interval opens Segment length (s) (1 to 3,600 seconds in 0.5-second steps, auto-defaulting around 1.5 to 15 seconds by track duration). Clicking Split audio cuts uniform sequential slices labeled Part 1, Part 2, and onward until the final frame.
Every extracted clip passes through applyCosineFadesInPlace, which applies Edge fade (ms) (0 ms to 250 ms in UI, 10 ms default) at both boundaries. Multiplying edge samples by 0.5 * (1 - cos(pi * i / N)) rounds abrupt cuts into smooth zero-crossing transitions.
Comparing silence, interval, and chapter split modes
Choosing between silence detection versus fixed interval and chapter splitting depends on whether your recording contains clear vocal pauses, continuous music, or published timestamps. Exported segments write standard 16-bit RIFF headers matching the Library of Congress Broadcast WAVE specification so desktop DAWs read track durations accurately.
| Split mode | Primary control | UI range & default | Worker boundary logic | Best workflow fit |
|---|---|---|---|---|
| By silence | Silence threshold (dB) | -60 to -15 dB (-38 dB) |
Flags 20 ms RMS windows under threshold |
Interviews, dictation notes, voice takes |
| By silence | Minimum pause (ms) | 150 to 3,000 ms (400 ms) |
Splits padded gaps (80 ms margins) meeting minimum |
Ignoring short mid-sentence word breaths |
| Fixed interval | Segment length (s) | 1 to 3,600 seconds (15) |
Slices sequential fixed-sample chunks | Long field captures, loop extraction, splits |
| Chapter timestamps | Chapter timestamps (one per line) | Multiline cue text | Parses MM:SS or HH:MM:SS intervals (>= 20 ms) |
YouTube chapters, albums, panel shows |
| All three modes | Edge fade (ms) | 0 to 250 ms (10 ms) |
Applies raised-cosine fade-in and fade-out | Preventing speaker clicks on sliced tracks |
Because speakers or songs recorded in different rooms often vary widely in volume, you can pass exported WAV files through Audio Normalizer afterward. Leveling each standalone track to -14 LUFS or -16 LUFS ensures consistent playback loudness across playlists.
Tip: Set Edge fade (ms) to
0 msonly when slicing beat-locked instrument loops right on zero crossings; for spoken recordings, interviews, and album chapters, keeping10 msto25 mseliminates click artifacts without softening initial consonants.
Exporting individual WAV tracks and batch ZIP archives
Once processing finishes, the results card reports extracted clip count, original track duration, and ZIP archive size. Reviewing each generated slice directly in browser memory lets you verify exact cut points before saving files locally.
Auditioning segments and bundling into ZIP locally
Every segment row displays #index, segment title, start and end timestamps, clip length, and byte size beside a Download .wav button and inline <audio controls> player. You can divide WAV files into individual 16-bit PCM tracks without re-encoding artifacts.
Clicking Download all segments (.zip) saves {stem}-segments.zip assembled in worker RAM via fflate zipSync at Deflate level 6 under the IANA application/zip media type registration. Zero-padded filenames such as 01-intro.wav keep multi-track folders sorted properly across operating systems.
Keeping multi-track exports private inside browser RAM
Client dictations, pre-release demos, and legal depositions remain inside your browser tab from initial drop through final ZIP download. Neither decodeAudioData, RMS silence scanning, nor zipFiles sends audio bytes, filenames, or timestamps over any network connection.
Workflows preparing WAV stem bundles alongside Unix backups or developer builds can also switch to POSIX TAR and GZIP archive utilities within the same local suite. Object URLs backing inline preview players revoke automatically when you re-run a split or close the tab.