Layering voice recordings and background music tracks locally
Combine spoken narration with three background music beds offline instead of opening a full desktop workstation. Keeping decoding, timing offsets, and track summing inside browser RAM ensures unreleased podcast interviews and client voice-overs never leave your device.
Adding a primary voice track and up to three bed stems
Dropping an audio file onto Drop a voice or podcast track here or clicking Load sample voice & music loads a 4.5-second speech take alongside a stereo chord pad. If you still need a clean vocal take before mixing, record clean voice narration directly in your browser and load that recording first.
Voice gain adjusts vocal loudness from -24 dB to +18 dB in 0.5 dB steps before beds sum into the stereo bus. Clicking + Add bed track loads up to three layers with independent Gain (dB), Offset (ms), Fade in (ms), Fade out (ms), and Loop bed controls.
Resampling mixed sample rates with Catmull-Rom interpolation
Compressed MP3, M4A, OGG, FLAC, and WebM stems decode locally through the MDN BaseAudioContext decodeAudioData reference, whereas RIFF WAV files parse directly in JavaScript. Mixed sample rates such as 22,050 Hz voice memos and 44,100 Hz music beds cannot sum sample-for-sample without pitch errors.
Each track runs through Catmull-Rom cubic interpolation inside our audio mixer to match the selected 48,000 Hz or 44,100 Hz stereo clock. Whenever a source stem downsamples below 90% of its original rate, a low-pass anti-aliasing filter at 0.44 times the target rate removes high-frequency foldback before summing.
How does automatic voice ducking keep music balanced?
Every 20 ms, the worker calculates root-mean-square energy on your voice stem and drops background beds whenever speech crosses 0.022 RMS amplitude. A 45 ms attack pulls music down rapidly on the first syllable, while a 380 ms release lets beds swell back gently between sentences without pumping.
Measuring 20 ms voice envelopes and attack or release curves
Envelope scanning and multi-track summing execute inside everyday.worker.ts using the MDN Web Workers API documentation so UI controls remain responsive on multi-minute podcasts. Using 20 ms evaluation blocks balances rapid consonant tracking against stable gain smoothing without jittering on tiny breath clicks.
If your microphone track carries heavy air-conditioner hum, background noise can falsely trigger ducking during silent pauses. Cleaning loud room hiss first with Audio Noise Remover keeps voice activity detection locked tightly onto actual spoken phrases rather than ambient fan noise.
Setting duck depth from -24 dB to -6 dB cleanly
Checking Auto-duck music during voice activates Duck depth, adjustable between -24 dB and -6 dB with a default -13 dB attenuation. Lighter values around -8 dB suit lively music beds behind casual vlogs, whereas deeper -16 dB to -20 dB cuts push busy instrumental tracks far beneath dense narration.
Each repeating loop boundary when Loop bed is active also receives a 12 ms raised-cosine micro-fade so looped percussion hits never click. Timeline length defaults to your primary voice track, or the longest offset bed duration when blending instrumentals without speech.
Audio mixer controls for gain, offset, fades, and ducking
Understanding how mixer controls interact with gain, offset, fades, and ducking prevents muddy overlaps before clicking Mix tracks. Exported uncompressed stereo buffers conform to the RIFF container architecture described in the Library of Congress Broadcast WAVE format description for predictable DAW alignment.
| Control | Range or options | Default value | Worker DSP stage | Practical mixing role |
|---|---|---|---|---|
| Voice gain | -24 dB to +18 dB (0.5 dB step) |
0 dB |
Linear gain multiplier on stereo voice | Balances quiet speech against loud music |
| Gain (dB) | -36 dB to +6 dB (0.5 dB step) |
-6 dB |
Per-bed attenuation before envelope | Sets base music or SFX bed loudness |
| Offset (ms) | 0 ms to 120,000 ms (50 ms step) |
0 ms |
Shifts bed placement on timeline | Staggers intro stingers or transition hits |
| Fade in (ms) & Fade out (ms) | 0 ms to 5,000 ms (50 ms step) |
150 ms / 300 ms |
Raised-cosine ramp on bed edges | Smooths music entries and tail exits |
| Auto-duck music during voice | Enabled / Disabled (-24 dB to -6 dB) |
Checked (-13 dB) |
20 ms RMS sidechain envelope follower |
Dips all beds smoothly during speech |
When you want chapters, jingles, and spoken segments placed strictly end to end instead of layered underneath narration, switch to sequential clip merging with cue sheets. That workflow preserves clean chapter markers rather than overlapping stems on a single timeline.
Tip: Set Gain (dB) near
-10 dBto-14 dBfirst, then let Duck depth handle another-10 dBto-13 dBattenuation during speech so music stays warm in pauses without startling listeners.
Exporting stereo WAV, M4A, and WebM mixes locally
After rendering, the result panel reports exact Duration, Sample rate, Peak level in dBFS, and output File size in bytes alongside a 160-bar SVG Mixed audio waveform. Every byte stays inside client memory throughout preview playback and file download.
Preventing distortion with soft-knee peak limiting
Summing a full-scale voice track with three musical beds can exceed 1.0 linear amplitude during loud choral peaks. Rather than hard-clipping values above full scale, softLimitSample keeps signals linear up to 0.85 amplitude (-1.4 dBFS) and smoothly curves peaks above 0.85 toward a 0.99 ceiling.
That single-pass soft-knee stage preserves vocal punch while keeping Download mixed WAV clean at 16 bits per sample. Watch Peak level after hitting Mix tracks; readings near -0.5 dBFS to -1.0 dBFS confirm strong loudness without harsh square-wave flattening.
Choosing 48,000 Hz or 44,100 Hz for podcasts and video
Switching Sample rate tailors soft-knee peak limiting at 44100 Hz versus 48000 Hz stereo clocks for your target platform. Select 48000 Hz when pairing dialogue with video timelines, or 44100 Hz when delivering audio-only podcast feeds or music masters.
Where your browser implements the W3C WebCodecs AudioEncoder specification, Download .m4a (128 kbps AAC) and Download .webm (128 kbps Opus) appear alongside WAV export. Creators mixing narration against video clips can also pull clean soundtrack beds out of MP4 clips locally before layering.