Player & audio editing

ChatGPT can play tracks and transitions to you in the preview player and render edited copies: shortened, tempo-adjusted, transposed or normalised. The original always stays unchanged.

ProLast updated:

On this page

Control the playerLink to “Control the player”

With the switch “Assistants may control the player” under Settings → Integrations → “ChatGPT & MCP” (off by default), an assistant may operate the preview player. The “Read library” permission is sufficient for this; without the switch, Sonodex declines any control.

ToolWhat it does
get_playback_statusLoaded track, position, duration, whether it is playing, volume and whether assistants are currently allowed to control. Also works without the switch.
load_trackLoads a track paused into the player, at a position if you wish
play_track, pause_trackStarts or pauses playback
seek_trackJumps to a position in the loaded track
set_preview_volumeChanges the player volume
audition_transitionPlays the end of one track from its mix-out range and then the start of the next from its mix-in range, 4 to 60 seconds each (default 16), one after the other without a crossfade
  • Sonodex visibly shows every control action: “AI assistant: … is controlling the player” with the action. “Stop” halts playback immediately and takes control away from assistants: the switch “Assistants may control the player” turns off, and further commands are declined. With “Allow again” in the notice you give it back.
  • Locked while a dialogue is open in Sonodex.
  • At most 30 control commands per minute per assistant. Nothing is saved.
  • audition_transition plays both excerpts one after the other, without a crossfade.

Edit audioLink to “Edit audio”

ToolWhat it renders
trim_audioSection from–to, with fade-in and fade-out (up to 30 s); cuts snap to beats of the stored beatgrid by default
adjust_audio_tempoNew tempo (target BPM or factor 0.8 to 1.25), pitch is kept; with Rubberband, otherwise FFmpeg atempo
transpose_audioPitch by −6 to +6 semitones, tempo is kept; without Rubberband via resampling, with a note about the lower quality
normalize_audioLoudness according to EBU R128 in two passes to −14 to −6 LUFS, true peak at most −1 dBTP (adjustable)
render_audio_editSeveral of these steps in one copy, always in the order cut, tempo, pitch, loudness
  1. The assistant calls one of the tools. This does not change anything: Sonodex renders a new file in the background; the assistant queries the progress with get_render_job and cancels its own render with cancel_render_job (without asking; the partial file is deleted).
  2. Once the render is finished, the confirmation dialogue appears with the original and the new file: format, length, BPM, key, target, the measured loudness and “Listen to result”.
  3. “Apply changes” puts the file next to the original or into your library folder, with a suffix in the name such as “Title (Intro Edit).flac”, and imports it as a new track, linked to the original. Tags and cover are carried over, BPM and key are adjusted. “Cancel” deletes the rendered file; Undo removes the new track and the file again.
  • Format: lossless sources (FLAC, WAV, AIFF, ALAC, WavPack) keep their codec; lossy ones become 24-bit FLAC so that no second lossy encoding is added. MP3 at 320 kbit/s is only available on request.
  • Changing BPM in the tag is not time-stretching. A new BPM value via the metadata does not change the audio; only adjust_audio_tempo actually renders it at the new tempo.
  • Limits: at most 2 renders per assistant and 6 in total, one running at a time; a render is cancelled after 10 minutes, and the result may be at most 60 minutes long.

StemsLink to “Stems”

Sonodex does not offer stem separation. separate_audio_stems and export_audio_stems always answer with unsupported_operation (reason stem_engine_missing). get_available_stems only lists stem files that already sit next to the track or in a folder with its name, and whether the file itself is a stem file (.stem.mp4).

SharesLink to “Shares”

TaskPermissionSwitch in Sonodex
Player status, render status, existing stemsRead library–
Control the playerRead libraryAssistants may control the player
Edit audio, cancel your own renderEdit tracksAssistants may propose changes

ExamplesLink to “Examples”

  • “Play me the transition from ‘Velvet Room’ into ‘Coastline’.”
  • “Make me an intro edit of ‘Glasshouse’: the first 32 bars, with a short fade-out.”
  • “Bring ‘Coastline’ to 126 BPM and normalise it to −9 LUFS.”

Something missing or wrong? Let us know. The app interface is available in English and German.

Original Sonodex app · Demo library