Perun Audio Editor

Waveform, playhead, zoom, pan, range selection, Trim, Split, Delete, Silence and cuts near zero crossings.

Cut, copy and paste fragments, duplicate ranges, insert silence and rebuild the structure of a recording.

Crossfade plays two neighbouring fragments at the same time across the transition instead of only sequential Fade Out/Fade In.

Move, remove, duplicate and trim segments, use named markers and export selected parts as separate files.

Build smooth crossfaded loops and extend background material to a chosen project duration.

Undo/Redo, autosave, session recovery and persistent history protect work during long projects.

Add independent sources and tracks, set Mute, Solo, level and pan, and move clips on the project clock.

Correlation analysis helps align two independent recordings that share common audio and shows a proposal before applying it.

Multiple time anchors separate a fixed offset from linear clock drift. An accepted correction creates a new track and never overwrites the original.

Background audio can be reduced automatically while a voice track is active, using a smooth gain envelope.

Choose a confirmed speech-free sample and fill a selected range with a non-destructive loop using short transitions and level control.

A locally imported ONNX model can prepare separate vocal and accompaniment results and add them as independent tracks.

Important — AI vocal separation: the current ONNX model requires at least 4 GiB of total RAM visible to Android and at least 1 GiB of available RAM before starting. A phone advertised as 4 GB may fall below this threshold because the system reserves memory. Close other apps if available RAM is too low. This requirement applies to AI separation, not to the entire editor. A compatible ONNX model must be imported separately. Processing runs locally; results and speed depend on the recording and device.

The app checks total and available RAM before starting AI vocal separation. If requirements are not met, it blocks this function and explains why. Other editor functions remain available. Closing other apps can help when available RAM is low, but cannot increase the phone’s physical RAM.

Monitoring modes help check the mix in mono and audition selected frequency ranges without changing the final export.

Repeat a selected range a chosen number of times, gradually change tempo and add a response pause.

Noise Reduction, 50/60 Hz hum filtering, Noise Gate, De-click and De-pop plus a prepared speech cleanup recipe.

De-clip attempts to soften local clipped regions; it does not claim to recover arbitrarily damaged recordings.

Mark or draw time-frequency regions to locally attenuate unwanted content in the frequency domain.

Protected regions can be masked in the shared preview and export path so raw audio does not return through ordinary playback fallback.

Equalization, high-pass and low-pass filters, plus approximate tone matching to a reference.

Compression, sample limiting, level control, mute, ReplayGain and BS.1770-based Auto Loudness.

Reverb, echo/delay, chorus, distortion, Reverse, pan, L/R swap and forced mono.

Envelope points on the timeline let level change smoothly within segments.

Automate Echo and Chorus mix over time without resetting their DSP history.

Phrase search and word-range selection can map directly to the matching area of the timeline.

A local two-model ONNX pipeline can segment speakers and assign clusters to a transcript. It is not identity recognition.

Choose an installed TTS engine, language and voice, set rate and pitch, render audio and add it as a project or track.

Silence analysis proposes a list of cuts before applying them instead of silently rewriting the recording.

Save personal processing recipes with icons, rename them and update them for reuse with other files.

Save independent project variants and return to earlier versions without overwriting alternatives.

Export a project together with referenced media and presets to a ZIP package, then import it as an independent project.

Checks missing sources, long silence, sample level, rendered duration, LUFS and stereo correlation/mono-loss risk, with navigation to reported areas.

Save and move sequences of up to 16 operations: pause shortening, cleanup, LUFS and final export. Batch runs start from freshly selected files and never inherit old URIs.

Export named markers to versioned JSON or readable TXT with timestamps mapped to the requested export.

A persistent queue supports progress, cancellation, retry and recovery after interruption.

Save the full mix, separate segments or selected tracks depending on the project.

Editing and rendering run on the device. The editor does not require an account for normal file-based work.