Try Pro FREE for 2 weeksGet Started
Start for free

audio-mix — Traditional AI Model

Audio Mix lays narration lines, background music and sound effects on one timeline and mixes them into a single sound. Narration lines follow one another with a gap you set, effects can be pinned to a line ("a second after line 3 ends"), the music is levelled under the voice and dips while anyone speaks, and the result is mastered to -14 LUFS. Connect a video to make the mix exactly as long as it. Put the result on the video with Audio Replace. CPU.

Run audio-mixBuild with it in Studio

Details

  • CategoryTraditional
  • Year-1
  • License

Inputs & Outputs

VoiceInputlist:soundoptional

Narration lines, in order (from Sound Collect)

MusicInputsoundoptional

Background music

EffectsInputlist:soundoptional

Sound effects, placed by effect_starts

VideoInputvideooptional

Only its length is used: the mix matches it

TimelineInputstringoptional

Placement as JSON; overrides the parameters

SoundOutputsound

The mixed track (WAV)

TimelineOutputstring

Where everything landed, as JSON

Parameters

gap_secfloatdefault 0.3range 0–10

Seconds of silence between narration lines.

lead_in_secfloatdefault 0range 0–60

Seconds before the first narration line.

voice_startsstringdefault advanced

Pin narration lines: positions separated by commas, blank = automatic. Positions are 0-based: a number is seconds; @3 / @3.end is voice line 3's start / end, shifted with +x or -x (e.g. @3.end-0.1); end-1.2 is 1.2 s before the mix ends.

One entry per voice line, comma-separated; a blank entry keeps that line automatic (after the previous line + gap_sec). A line may anchor only on earlier lines (@0..@i-1). Positions are 0-based: a number is seconds; @3 / @3.end is voice line 3's start / end, shifted with +x or -x (e.g. @3.end-0.1); end-1.2 is 1.2 s before the mix ends.

effect_startsstringdefault

Where each effect starts, one per effect, separated by commas. Positions are 0-based: a number is seconds; @3 / @3.end is voice line 3's start / end, shifted with +x or -x (e.g. @3.end-0.1); end-1.2 is 1.2 s before the mix ends.

Required when Effects is connected: exactly one position per effect, comma-separated, in list order. Positions are 0-based: a number is seconds; @3 / @3.end is voice line 3's start / end, shifted with +x or -x (e.g. @3.end-0.1); end-1.2 is 1.2 s before the mix ends. Missing or extra entries fail the run.

music_volumefloatdefault 1range 0–2

Music level on top of the automatic bed level. 1 keeps it, 0 mutes it.

duckingstringdefault mediumoptions: off, light, medium, strong

How far the music dips while narration plays: light -6 dB, medium -12 dB, strong -18 dB.

'medium' (default) = -12 dB, 'light' -6, 'strong' -18, 'off' none. The dip covers each placed voice line exactly, with 150 ms ramps, whatever the line's own loudness.

effect_volumefloatdefault 1range 0–2

Volume of the sound effects. 1 keeps them as they are.

lengthstringdefault autooptions: auto, video, voice, music, fixed

What sets the mix length. Auto: the video if connected, else the narration and effects, else the music.

'auto' (default): the Video input's length if connected, else the last voice line or effect + tail_sec, else the music's length. 'video' / 'voice' (+ tail_sec) / 'music' force one; 'fixed' uses duration_sec. At most 600 s; a line or effect that would start after the end fails the run.

duration_secfloatdefault 0range 0–600advanced

Mix length in seconds when length is fixed.

tail_secfloatdefault 0.5range 0–10advanced

Seconds kept after the last narration line or effect.

voice_levelstringdefault matchoptions: match, as-isadvanced

Match levels every narration line to the same loudness; as-is keeps them.

music_fitstringdefault loopoptions: loop, onceadvanced

Music shorter than the mix: loop it, or play it once.

music_fade_out_secfloatdefault 1range 0–10advanced

Fade-out at the end when the music is cut.

target_lufsfloatdefault -14range -31–-9advanced

Target integrated loudness in LUFS. -14 is the social/YouTube norm.

true_peakfloatdefault -1.5range -9–0advanced

True-peak ceiling in dBTP.

duck_onstringdefault lineoptions: line, speechadvanced

Line: dip the music for each whole narration line. Speech: only while someone is actually talking inside it.

'line' (default): the music dips over each placed voice item from start to end. 'speech': only over the talking inside each item - a stretch 20 dB under the item's own loudness for 0.3 s+ is a pause, and pauses under 0.5 s are bridged so the music does not pump between words. Use speech for a voice that is mostly not talking, such as dialogue taken out of a video with Video Audio Extract.

effect_volumesstringdefault advanced

Volume of each effect, comma-separated in list order (0-2, 1 keeps it). A blank or missing entry uses Effects volume.

Optional per-effect linear volume, comma-separated in list order, each 0-2 (2 = +6 dB). A blank or missing entry uses effect_volume; more entries than effects fail the run. Use it to sit an ambience bed under short accents, e.g. '0.6, 1, 1'.

effect_widthsstringdefault advanced

Stereo width of each effect, comma-separated in list order (0-1, 0 keeps it as it is). Spreads a mono bed such as rain or wind without changing how it sounds on one speaker.

Optional per-effect stereo width, comma-separated in list order, each 0-1 (default 0 = unchanged). Adds decorrelated side around the untouched mid, so mono playback stays the dry sound. Generated effects are mono; 0.6-0.8 suits an ambience bed (rain, wind, crowd), keep short accents at 0. More entries than effects fail the run.

effect_fade_out_secfloatdefault 0.5range 0–10advanced

Fade-out for an effect that the end of the mix cuts short. 0 cuts it dead.

Seconds of fade-out applied to any effect that is still playing when the mix ends (default 0.5; 0 = hard cut). Effects that end inside the mix are untouched.

Tags

  • audio
  • mix
  • ducking
  • narration
  • music
  • sfx
  • loudness
  • timeline

Alternatives in Video Editing

All AI models