Keep auditable frame maps when editing recordings
Treat pause cuts, channel recovery, full-spectrum review and filters as explicit handoffs with complete evidence.
Start with the actual defect
A class-recording question asks to remove long pauses; another user needs a useful left channel duplicated because the other side is noisy. A developer wants a spectrum for the full file rather than its currently playing buffer. These are different tasks. Record which observation led to which operation, and keep the original recording.
Preserve timing and channel identity
For pause shortening, record threshold, window size, minimum run and retained duration. The checked10s recording becomes7.2s by removing22,400 source frames. A complete half-open frame map explains which output span still corresponds to which source span. A timestamp from the original cannot be used on the edited copy without that map. Quiet content is not automatically useless; listen around cuts.
For channel recovery, mapping1,1 duplicates channel1 without mixing. Opposite-phase channels can cancel if averaged; explicit selection avoids that unrequested operation. Split-all preserves each channel as its own mono WAV, with source indexes and statistics. PCM16 selection/cuts retain actual integer bytes; MP3 uses browser-decoded samples and a declared PCM16 quantizer.
| Evidence | Question answered | Limit |
|---|---|---|
| Original/output frame map | Where did retained audio come from? | No judgement that removed low-level content was meaningless |
| Source channel indexes and byte-exact PCM16 comparison | Which source channel supplied each output? | No automatic speaker or channel-role inference |
| Full matrix, axes and settings | What is present across the entire analysed channel? | Window resolution and right-padding remain relevant |
| Declared sections, poles, response and clipping | What transformation was applied? | Frequency bands may contain useful material; high Q can ring |
Keep analysis complete and budgeted
A labelled preview helps navigate, but the full matrix and settings support later measurement. The4s/8kHz example has61,500 amplitudes, not just the columns visible in its PNG. Peak values are retained for every window; ties are explicit. Larger inputs exceeding4 million cells or100 million FFT work units are rejected with a recovery path rather than reduced to a short excerpt.
The notch example measures60Hz suppression while retaining a1kHz component. W3C/RBJ coefficients, stable poles,512 response points, zero-state processing and reported clipping describe that exact operation. They do not turn frequency filtering into speech recognition or general noise removal. Use a narrower chain only when the recording supports it.
Make decoder differences visible
The same local MP3 fixture decodes to2.000s in Chromium and2.040s in WebKit. Priming and gapless handling change PCM frame counts; the report names the actual rate, frames and timeline. The uniform output rule is round(decoded×32768), clamped to signed16. Comparing sample counts across browsers without the decoder context can misdescribe a cut.
- Keep source plus all downloads; a200-row preview is a navigation aid.
- Play controlled generated WAV previews manually, and listen at removed boundaries and clipped peaks.
- Record operation parameters and source-channel meaning yourself; the tool does not infer them.
- Use WAV PCM16 for exact source-byte comparisons. For other depths or MP3, compare against actual decoded PCM and declared quantization.
- Treat failed validation or complexity limits as a rejected task; correct the input or choose another sample and rerun.
References
- User task: shortening long pauses in class recordings
Question body read; low-level pause removal is distinct from removing continuous background noise.
- User task: assign the useful left channel to both outputs
Question body read; explicit channel choice is supported without guessing which channel contains speech.
- Developer task: full-file spectrogram instead of current playback buffer
Question body read; complete analysis requires a matrix and full timeline, not a moving display.
- User task: frequency-specific processing of a poor voice recording
Question body read; the hum and low-frequency components match explicit filters; clicks and complete denoising are outside this tool.
- W3C Audio EQ Cookbook
Normalized biquad coefficients and response definitions; technical reference, not independent demand evidence.
Tools in this category
Expand a tool to see its steps, options and supported formats, then open its workspace.
Shorten low-level recording pausesShorten long pauses under an all-channel RMS threshold and export complete retained-frame mappings.
Inspect rounded20ms windows of actual PCM, requiring every channel to be under the threshold. Remove the middle of qualifying pauses while retaining both edges. Download an actual PCM16 WAV and complete quiet/deleted intervals and half-open source/output frame mappings. Retained PCM16 frames are byte-exact without crossfade. This is not speech recognition; useful quiet sounds may be removed.
Steps
- Choose local WAV/MP3.
- Set parameters and inspect actual results,statistics or frame mappings.
- Listen/inspect and download complete outputs.
Available options
- RMS threshold (dBFS)
- -40
- Minimum pause (ms)
- 500
- Retained pause (ms)
- 200
Must be≤minimum; split retained frames as floor-left/ceil-right.
Capabilities and limits
- One supported WAV/MP3 up to30 MiB and80 MiB actual decoded PCM;1–8 channels,8–192kHz; combined downloads100 MiB. WAV uses existing PCM8/16/24/32 and float32/64 support; MP3 must actually decode in the browser. No microphone, upload or remote audio fetch.
- MP3 priming/gapless behavior varies by browser: the same checked MP3 produced2.000s in Chrome and2.040s in WebKit. All times/frames follow this browser’s actual decoded PCM; cross-browser sample/duration identity is not guaranteed. PCM16 output rounds decoded*32768 and clamps[-32768,32767] without normalization; float/24/32/MP3 original encoding bytes are not preserved.
- Threshold−80…−10dBFS; minimum pause100…60,000ms; retained pause20ms up to minimum;10,000 pauses. Threshold comparison is inclusive; the final short window uses its actual samples. Long all-quiet input retains selected edges; input shorter than the minimum is not cut. Zero detections still deliver the original frame sequence.
Recover and split recorded channelsExplicitly isolate, repeat or split existing source channels into playable WAV files with full mappings.
Inspect per-channel RMS, peak, DC and over-full-scale samples, then choose1-based source indexes. Mapping1,1 repeats the left channel on both sides without averaging in right-channel noise. Split every source channel into a ZIP with individual playback. PCM16 selected/repeated/split samples are byte-exact. Channel meaning is not guessed and missing independent stereo cannot be recreated.
Steps
- Choose local WAV/MP3.
- Set parameters and inspect actual results,statistics or frame mappings.
- Listen/inspect and download complete outputs.
Available options
- Delivery mode
- Explicit mapping · Split all source channels
- Source channel list (1-based)
- 1,1
1,1 repeats channel1;2,1 swaps stereo;1 isolates channel1. Ignored in split mode.
Capabilities and limits
- One supported WAV/MP3 up to30 MiB and80 MiB actual decoded PCM;1–8 channels,8–192kHz; combined downloads100 MiB. WAV uses existing PCM8/16/24/32 and float32/64 support; MP3 must actually decode in the browser. No microphone, upload or remote audio fetch.
- MP3 priming/gapless behavior varies by browser: the same checked MP3 produced2.000s in Chrome and2.040s in WebKit. All times/frames follow this browser’s actual decoded PCM; cross-browser sample/duration identity is not guaranteed. PCM16 output rounds decoded*32768 and clamps[-32768,32767] without normalization; float/24/32/MP3 original encoding bytes are not preserved.
- Mapping is a comma list of1–8 output channels and every index must exist; split mode ignores the mapping. Output capacity is checked before allocation; channels are never silently dropped.
Full-recording spectrogram and numeric exportAnalyze the complete recording and export a labelled PNG, full STFT amplitude matrix and explicit window/time/frequency settings.
Select an actual source channel and compute the full periodic-Hann STFT. Download PNG with t[s],f[Hz] and dBFS colour scale, every little-endian float32 amplitude bin, actual window-centre times, peaks and settings. Display may max-pool within image budgets while covering every window; complete numeric data remains downloadable. No recording prefix is silently selected.
Steps
- Choose local WAV/MP3.
- Set parameters and inspect actual results,statistics or frame mappings.
- Listen/inspect and download complete outputs.
Available options
- Source channel (1-based)
- 1
- FFT window samples
- 256 · 512 · 1024 · 2048 · 4096
- Display dBFS floor
- -80
Capabilities and limits
- One supported WAV/MP3 up to30 MiB and80 MiB actual decoded PCM;1–8 channels,8–192kHz; combined downloads100 MiB. WAV uses existing PCM8/16/24/32 and float32/64 support; MP3 must actually decode in the browser. No microphone, upload or remote audio fetch.
- MP3 priming/gapless behavior varies by browser: the same checked MP3 produced2.000s in Chrome and2.040s in WebKit. All times/frames follow this browser’s actual decoded PCM; cross-browser sample/duration identity is not guaranteed. PCM16 output rounds decoded*32768 and clamps[-32768,32767] without normalization; float/24/32/MP3 original encoding bytes are not preserved.
- FFT power-of-two256…4096,hop=FFT/4,no left padding and a zero-padded final window. One-sided amplitude=absFFT/HannSum, doubled except DC/Nyquist. Up to4,000,000 cells and100,000,000 FFT work units, checked before computation. PNG spans full input time, assigns columns by window centres with nearest-window fallback for empty columns, and covers0…Nyquist. Peak bins are amplitude maxima; exact ties choose the lowest bin and report count. This is not pitch or source certification.
Apply explicit recording frequency filtersApply explicit high-pass, low-pass or notch chains and export WAV, response and clipping counts.
Enter type,frequencyHz,Q per line, such as notch,60,10. Apply the same explicit chain to all channels with stable W3C/RBJ biquads and zero initial state. Download an actual PCM16 WAV with complete coefficients,poles,response and before/after levels. No normalization or automatic denoise; high Q can ring, remove useful sounds or cause clipping.
Steps
- Choose local WAV/MP3.
- Set parameters and inspect actual results,statistics or frame mappings.
- Listen/inspect and download complete outputs.
Available options
- Filter chain: type,Hz,Q per line
- notch,60,10
Use highpass,lowpass,notch; up to8 lines,frequencies strictly below Nyquist.
Capabilities and limits
- One supported WAV/MP3 up to30 MiB and80 MiB actual decoded PCM;1–8 channels,8–192kHz; combined downloads100 MiB. WAV uses existing PCM8/16/24/32 and float32/64 support; MP3 must actually decode in the browser. No microphone, upload or remote audio fetch.
- MP3 priming/gapless behavior varies by browser: the same checked MP3 produced2.000s in Chrome and2.040s in WebKit. All times/frames follow this browser’s actual decoded PCM; cross-browser sample/duration identity is not guaranteed. PCM16 output rounds decoded*32768 and clamps[-32768,32767] without normalization; float/24/32/MP3 original encoding bytes are not preserved.
- Up to8 sections and160,000,000 sample-section work units; frequency10Hz…0.49sampleRate,Q0.1…30,finite coefficients and strictly stable poles. Float64 samples/state between sections, final float32 PCM then explicit quantization; samples beyond±1 count as clippedSamples and WAV saturates visibly in the report. Response SVG displays−100…20dB with all512 numeric response points. No speech recognition or full noise restoration.