Skip to content
HyperPowers
04 · HyperAudio

Clean the voice, split the music,cut silences on the waveform.

With AI that runs on your computer, no cloud. Clean the voice first; separate stems if you need them.

Clean voice firstLocal AI, no cloudStems: macOS todayWaveform previewUndo with Reset
Selected clipCleanup & stems
MixVoiceNoisy voice
InstrumentalDrumsClean voice
Bass
Other

Illustrative animation. Pause it with the button at the bottom right.

The problem.

Sound familiar?The interview was recorded in a café.

Music was playing, or an air conditioner was humming next to the mic. Or the client wants the song without vocals, or just the drums for an intro remix.

HyperAudio separates and cleans the audio of the selected clips with AI tools installed on your computer, and cuts silences on the waveform with a preview before you touch the timeline.

What it does.

Two cards: Cleanup & stems, and the Silence remover.

Clean voice

The first option and the default. Removes background noise without separating anything and leaves the clean voice on a new track.

Stems

Vocals + instrumental, vocals only, 4 stems (vocals, drums, bass, other) or 6 (adds guitar and piano). New tracks go under the original, which is muted. Reset brings it back.

Vocals + instrumentalVocals only4 stems6 stems

Options

Maximum quality: cleaner, about four times slower. Mute the original. Clean the voice on the vocal track as well.

Local, no cloud

Demucs (from Meta, MIT license) separates and DeepFilterNet cleans. You install the engine once from the card, about 1 GB. Stems work on macOS today.

Silence remover

Analyzes the track you choose, on your computer, and draws the waveform with what stays and what goes. Detect by dB, or with AI: the local transcript refines where each cut starts and ends, taking breaths and voiceless noise with it.

Threshold, silence, padding

Threshold, Min. silence and Padding decide what counts as silence, from how long it gets cut and how much air is left on each side. Preview lets you listen first.

Two ways to cut

Cut only this track leaves gaps and moves nothing else. Jump cut (whole timeline) cuts every track and closes the gaps. Recover audio undoes the last cut.

Ask by typing

In HyperChat: “split the voice from the music in the selected clips” or “set the stems card to vocals only”.

Work flows.

Two jobs, with the buttons named as in the panel.

Workflow 1

The café interview

Clean the voice, and keep the original one click away.

  1. 01

    Select the clip

    Select the audio clip, or the video with its audio.

  2. 02

    Pick the mode

    Clean voice is the default. Choose Vocals only with Clean the voice if there is also music.

  3. 03

    Press the button

    The button reads Clean in cleaning mode and Split in the others. The result appears underneath; the original is muted. Reset brings the original back.

Workflow 2

Silences, with waveform precision

It cuts on the waveform of the track you choose, with a preview, so you decide exactly how much air stays.

  1. 01

    Analyze

    Choose the track and detect by dB or with AI. The waveform shows what stays and what goes.

  2. 02

    Tune and listen

    Adjust Threshold, Min. silence and Padding, then Preview.

  3. 03

    Cut

    Cut only this track to leave gaps, or Jump cut (whole timeline) for a talking head where everything follows the voice. Recover audio undoes it.

Audio track · A1Silence remover · dB
talk shorter
VoiceCut outAir left (padding)
The Silence remover card before analyzing: detection mode, threshold, minimum silence and padding controls.
Real panel screenshots, Premiere Pro 26.5, macOS.

Illustrative animation. Pause it with the button at the bottom right.

Good to know.

Stems are macOS only for now
The engine installs once, about 1 GB. It does not work on Windows yet.
AI detection needs Whisper
The local transcript refines the cuts. Whisper is installed separately for now.
Maximum quality is slow
It separates more cleanly and takes about four times longer.