Ambience Bed for Video and Audio

What it is for

A clip recorded in a quiet room, a screen recording or B-roll without sound often feels empty. An ambience bed — a soft, continuous background like rain, wind or room tone — glues cuts together and hides small noise jumps. This tool generates such a bed with the usefulweb audio engine, loops it seamlessly over the whole clip, fades it in and out, and mixes it under the original sound or replaces the sound with it.

Videos keep their picture untouched: the video stream is copied, only the sound is re-encoded (AAC in MP4, Opus in WebM). Audio files come back as a 16-bit WAV at 48 kHz. The beds are synthesized impressions, not recordings of real places, and the tool does not duck the bed under speech automatically — you choose the level.

It works on clips up to five minutes. Videos in codecs that cannot be copied into MP4 or WebM (for example WMV) need converting first.

Worked example

The first example puts light rain under a one-second test clip: Ambience Rain, Bed level −38 dBFS, fade-in 0.2 s, fade-out 0.3 s, no room reverb, Original sound kept. The tool returned an MP4 with the picture copied and:

• Bed level −38 dBFS RMS, original level −24.3 dBFS RMS.
• Bed under the original 13.7 dB — comfortably in the background.
• Mix peak −17.8 dBFS, no samples touched by the limiter.

Raising the bed to −20 dBFS with wind made the bed 4.3 dB louder than the original. The tool flagged it with a warning that the bed will cover speech; the mix peak rose to −7.5 dBFS. Switching to an evolving drone at −24 dBFS in a concert hall with Original sound set to replace returned the same picture with only the drone, so the original level was empty.

How to read the result

Bed level is the RMS of the generated loop before fades. Original level is the RMS of the clip's own sound, and Bed under the original is the difference: around 12–20 dB keeps a bed in the background under speech, below 6 dB it competes with the voice, and a negative number means the bed is louder.

Mix peak is the loudest sample of the new sound. If the sum comes within 0.2 dB of full scale, a soft limiter rounds off only those peaks and the number of affected samples is reported. loopSeconds is the length of the seamless loop that repeats under the clip (up to 20 s). The spectrum chart shows the new sound's frequency balance.

Tips and common mistakes

• Start around −32 to −38 dBFS for a bed under speech and adjust by ear; RMS says little about perceived loudness.
• Room tone with hum works well to hide cuts in dialogue; rain and stream add more movement.
• Fades longer than half the clip are shortened to half, so very short clips still fade in and out.
• Use the same seed for every clip of a series so the bed sounds identical across episodes.
• A room reverb makes the bed sound further away; a forest or small room is subtler than a cathedral.
• Normalise loudness after adding the bed, not before, if you deliver to a streaming target.

FAQ

Is the video re-encoded?

No. The picture stream is copied bit for bit. Only the sound is encoded again, as AAC 192 kb/s in MP4 or Opus 160 kb/s in WebM.

What happens with a video that has no sound?

The result contains only the bed, and the report says so with a warning. The original level and the difference stay empty.

Will the bed repeat audibly?

The loop is up to 20 seconds long and seamless; noise-like ambiences hide the repeat well, while a drone's slow movement can become recognisable on long clips.

Can I use the result commercially?

Yes. The ambiences are generated by usefulweb's own code without samples, so there are no third-party rights in the bed.