DEX 4.1 Beta: AI DJ Tools, Better Stems and Karaoke

DEX 4.1 beta is now available, and this is our biggest DEX 4 update yet. With a new playback engine, AI-powered track analysis, higher-quality stem separation, synced lyrics, and smarter track suggestions, this release brings substantial improvements to how you prepare your music and perform.

We’re inviting the PCDJ community to download the beta, explore the new features, and help us get DEX 4.1 ready for its official release. Whether you mix music, perform with video, host karaoke, or combine all three, there is plenty to test.

Alongside the new creative tools, DEX 4.1 includes improvements to audio performance, library protection, streaming, and day-to-day reliability. You’ll find the highlights and the complete release notes below.

What’s New in DEX 4.1?

The improvements in DEX 4.1 extend from the playback engine to the library, waveforms, streaming integrations, and karaoke workflow. Here are the highlights, followed by the complete release notes.

1. A New Time-Stretch and Key Lock Engine

Key lock and key shift now use our new engine when Time-stretch quality is set to High. The result is cleaner sound while beatmatching, improved bass handling, and lower CPU usage.

The new engine also eliminates warbling and brief detuning when you nudge or bend the tempo downward with key lock enabled. The Normal quality setting retains the classic engine.

2. AI Beat Detection

A new AI engine detects tempo, beats, and downbeats more reliably. Manually edited beat grids are preserved during routine reanalysis, so your preparation work stays intact.

A new Re-analyze the whole library action is also available in the track menu. Unlike routine reanalysis, this action overwrites manually edited beat grids after confirmation.

3. More Accurate Musical Key Detection

DEX 4.1 introduces a new musical key detector that is substantially more accurate than its predecessor. On a standard test set, it identified the correct key on 62% of tracks, compared with 37% for the previous detector.

Key detection is now always enabled.

4. Song Structure and Automatic Cue Points

DEX 4.1 analyzes tracks for sections such as intros, builds, drops, main sections, breaks, and outros. These sections appear as colored bands on the overview and in a dedicated lane on the deck waveform.

When a track has no existing hot cues, DEX can populate them automatically from the detected structure. Your existing cues are preserved, and Automix uses the detected mix-out point unless you set one yourself.

5. Track Energy and Mood

Tracks receive an energy level from 1 to 10, compatible with Mixed In Key tags, along with a mood label such as Happy or Calm.

Energy and mood appear in dedicated library columns and directly in modern track rows, such as “E7 Happy,” giving you additional context when choosing your next track.

6. Next Picks: Smarter Track Suggestions

The new Next Picks list suggests what to play after the track on the master deck. Recommendations consider key compatibility, tempo, energy, mood, similar sound, available mix room, and how recently you played a track.

Presets include Mixable, Same key, Same tempo, and Fresh. Each suggestion explains why it was selected, and you can right-click any track to access Find similar tracks.

7. Synced Lyrics

DEX 4.1 reads synced lyrics from LRC files and tags, supports online lookup through LRCLIB, and can use AI to align plain lyrics with the vocals.

Lyrics appear on the video output and along the deck waveform. You can also export an .lrc file alongside the track. A new Lyrics tab in Preferences brings the related settings together.

8. Karaoke Mode for Regular Songs

Karaoke mode uses AI stems to remove the vocals from an ordinary song while keeping key lock enabled. An optional guide vocal can remain at -18, -12, or -6 dB.

The mode switches on automatically when a singer’s ordinary track is loaded from the rotation list.

9. Better Stems and Updated AI Unmix EQ Defaults

A new Best quality stems option uses an improved separation model for noticeably cleaner vocals and instruments. It takes longer to process than the standard model and is disabled by default.

AI Unmix EQ is now enabled by default on new installations running on machines with at least 7 GB of RAM. Existing saved settings are respected, and stem adjustments no longer produce clicks.

10. RGB Waveforms

A new RGB waveform style colors the deck waveform and whole-track overview according to bass, mids, and highs.

Find it under Preferences > General > Waveform colors.

11. Smoother Sync

Sync now aligns the phase with a brief tempo bend rather than a jump. You can choose beat, bar, or phrase alignment.

Key sync also clears stale key shifts when the decks are already compatible.

12. Storage and Backup Controls

The new Storage & Backup tab lets you move DEX data to another folder or drive, manage automatic backups, and restore a previous backup with one click.

Database recovery has also been improved. For libraries with fewer than 100,000 tracks, changes are automatically saved every five minutes rather than waiting until you exit.

13. Automatic Microphone Feedback Suppression

An automatic feedback suppressor helps reduce howling on the On Air microphone.

The feature is enabled by default and can be managed in Audio settings.

14. Restored ShaderToy Visuals

You can add a ShaderToy shader by pasting its code or opening a local .frag file.

This provides an alternative to URL-based importing, which is affected by restrictions on the ShaderToy API.

15. Streaming Support for Filler Music and Preview

The Filler music player now supports third-party streaming tracks, and the Preview player does, too.

16. Major Stability and Performance Improvements

The audio engine has been reworked to address dropouts associated with loading, unloading, hot cues, and sync jumps.

Audio on 48 kHz devices previously played 0.34% too fast and is now exact. The release also significantly reduces idle CPU usage and addresses the Party Tyme catalog issue that caused DEX 4 to crash at startup.

Complete DEX 4.1 Beta Release Notes

Playback Engine

• Introduced a new time-stretch and pitch-shift engine for key lock and key shift when Time-stretch quality is set to High.
• Fixed warbling and brief detuning when nudging or bending the tempo downward with key lock enabled.
• Improved bass handling and transients with key lock enabled, keeping kick attacks sharp.
• Significantly reduced CPU usage for key lock. The new engine uses less CPU than the previous engine on both Mac and Windows.
• Hot cues, loop wraps, and sync jumps on a key-locked deck no longer cause callback peaks or dropouts at small buffer sizes.
• Loading a track onto a stopped deck and pressing Play no longer produces a CPU spike in the audio thread, reducing glitches on slower machines.
• Stem separation after loading a track no longer plays partially written stems. The deck continues playing the original mix until the stems are ready.
• Dropout and overload counters are maintained for the audio device, and the audio load meter reports true peaks.
• Effect tails decaying to silence no longer cause CPU spikes on Intel machines.
• A driver forcing a very large buffer, over approximately 100 ms, no longer crashes the application.
• Fader, crossfader, trim, and EQ changes are ramped by the audio thread, eliminating occasional clicks during adjustments.
• The three-band EQ is faster, and its band gains ramp evenly on both channels.
• Flanger and reverb: the right channel no longer plays a burst of old audio when the effect is switched on again.
• Audio recording: blocks lost due to a slow disk are counted and logged instead of being silently omitted.
• Files with sample rates other than 44.1 kHz, including 48 kHz tracks and most videos, are resampled using a high-quality filter instead of per-block linear interpolation.

Track Analysis

• A new AI beat, tempo, and downbeat engine replaces the previous beat detector. Tempo sections are created only where the tempo actually changes.
• Added automatic 3/4 versus 4/4 meter detection. The time signature is stored with the beat grid.
• Phrase detection, using four or eight bars, and the phrase anchor establish bar 1 of the grid so that bar counting aligns with the music.
• Song structure sections, including intro, build, drop, main, break, and outro, plus verse, chorus, and bridge labels, are detected across the library. Detection uses stems when AI Unmix EQ is enabled and the original mix otherwise.
• Vocal sections, distinguishing lead vocals from vocal textures, are detected from the vocal stem and displayed on the overview.
• Automatic cue points based on song structure populate the hot-cue slots when a track has none. Mix in cues are placed 32 or 16 bars before the first body section. Existing cues are never changed.
• Mix-in and mix-out points are stored per track. Automix uses the detected mix-out point unless you set one yourself.
• Introduced a significantly more accurate key detection engine that also reports tuning and confidence. The Detect Musical Key setting has been removed because key detection is now always enabled.
• Each track receives an energy level from 1 to 10. Values are read from Mixed In Key or Lexicon tags when available; otherwise, they are computed and calibrated against Mixed In Key labels. A new Energy column is available.
• Each track receives mood data, including valence, arousal, and a mood word, from a new AI mood model. A new Mood column is available.
• Energy and mood appear in modern track rows before the key, time, and BPM information. The Show energy and mood in rows preference is enabled by default.
• An energy curve is generated for each track. An energy strip on the position slider is available to skins and is disabled by default.
• Manually edited beat grids are preserved during reanalysis. Only Re-analyze the whole library overwrites them, after confirmation.
• Each track remembers which analysis version produced its data. Tracks analyzed by an older version are analyzed again when loaded or in the background, while preserving cues and user-edited grids.
• Added the track menu action Re-analyze the whole library, covering BPM, beat grid, meter, and key.
• Analysis models are loaded once and warmed up at startup so that the first analysis is no slower than subsequent analyses.

Stems and AI Unmix EQ

• The new Best quality stems option uses an improved separation model for noticeably better vocal and instrument separation. Processing is approximately five times slower than the standard model. This option is disabled by default.
• AI Unmix EQ is enabled by default for new installations on machines with at least 7 GB of RAM. Existing saved settings take priority. Stems options can still be enabled on machines below this threshold.
• Stem gain changes from pads and AI Unmix EQ knobs ramp over 10 ms, eliminating clicks.
• Momentary stem pads restore the exact previous gain when released.
• EQ knobs control stems only when the track actually has stems. Tracks without stems, including tracks longer than 15 minutes or tracks from a service that prohibits separation, use normal EQ. Stem gains from the previous track are no longer carried over.
• Pads 9 through 16 no longer trigger actions from the next pad mode when using the eight-pad modes: Roll, Slicer, Stems, and Sampler.
• Windows: CPU-based stem separation leaves two cores available for playback.

Lyrics and Karaoke

• Added support for synced lyrics from .lrc files alongside the track and from ID3 tags, including SYLT and USLT, with support for older iTunes-style line breaks.
• Added optional online lyrics lookup through LRCLIB.
• Plain, unsynced lyrics are aligned to the vocals using an AI aligner on the vocal stem when the track is loaded onto a deck. Lyrics that do not match the recording are rejected and remembered as not found.
• Lyrics appear on the video output and along the deck waveform, with a per-track timing offset.
• Track context menu > Lyrics provides options to look up lyrics, align them again, adjust timing, show the file, remove lyrics, and export an .lrc file alongside the track.
• A library marker identifies tracks with lyrics. A #lyrics search term is also available.
• Per-deck Karaoke mode removes vocals through AI stems while keeping key lock enabled. Toggle it from Stems pad 9, the track context menu, or a controller using deckKaraoke. It switches on automatically when a singer’s ordinary track is loaded from the rotation list.
• The guide vocal preference lets you remove vocals completely or retain them at -18, -12, or -6 dB.
• The lyrics overlay follows the playing deck in Karaoke mode, even when show on video is disabled.
• The Karaoke tab in Preferences is now split into Karaoke, covering singers, filler music, ticker, and rotation, and Lyrics, covering Karaoke mode and lyrics options.
• Party Tyme catalog: fixed the startup crash caused by the catalog format change of September 29, 2026.
• Load next singer no longer crashes when the rotation list is empty.
• Filler and background music from a streaming service is downloaded and buffered 30 seconds before the mix starts. The Preview player can also play streaming tracks.
• Fixed stale check marks in menus, including the Karaoke mode menu item.

Next Picks

• The new Next Picks special list appears above History and ranks the library against the master deck’s track using key compatibility, including fuzzy key shifts; tempo, including half-time and double-time; genre; energy; mood; similar sound; freshness; rating; and available mix room.
• Presets include Mixable, Same key, Same tempo, Fresh, Lift, Cool down, and Similar sound. Options include half/double tempo, hiding played tracks, and selecting the scope: the whole collection or the last browsed list.
• Each suggestion displays the reasons it was selected, such as “same key, mix room 32 bars.”
• Find similar tracks (Next Picks) is available in the track context menu.

Sync and Cues

• Sync aligns the phase with a brief tempo bend instead of a jump. This option is enabled by default.
• Sync phase mode can be set to beat, bar, or phrase, with bar as the default. Find it in Deck settings next to sync-lock.
• When a track’s bar 1 is uncertain, sync falls back to beat alignment instead of forcing an incorrect bar alignment.
• Key sync clears a stale key shift when the decks are already compatible. Strict mode now shifts by a maximum of two semitones, compared with up to six previously.
• Phrase skip jumps by the track’s detected phrase length instead of a fixed four bars.
• Hot cue button hints display the cue’s name.
• Automix: when the list runs out, the final track plays to the end instead of the last two tracks alternating indefinitely. Fade time no longer depends on the frame rate.

Waveforms and Display

• Added a new RGB waveform style for the deck waveform and whole-track overview. Find it under Preferences > General > Waveform colors, with options for As in the skin, Layers, and RGB.
• The overview on the position slider draws the same low/high layers as the deck waveform.
• Structure sections, vocal sections, and the song-structure lane are drawn on the overview and deck waveform.
• The waveform, cue markers, lyrics, jog wheel, and video now display the position you are actually hearing, with device latency compensation. Scrolling is smooth rather than limited to whole-pixel increments.
• Beat and downbeat lines retain their own colors. Bar lines no longer wobble by a pixel while scrolling on non-Retina displays.
• Hints, menus, and lyrics are drawn above waveform lines. Soft text edges no longer have dark fringes.
• Idle rendering: when nothing is playing or moving, the window redraws at approximately 15 fps. It does not redraw while minimized or hidden. Idle CPU usage dropped from approximately 16% to 2%. A selected shader visual maintains the full frame rate.
• Render thread CPU usage with lyrics and a visual enabled has been reduced by more than half.

Library and Playlists

• For libraries with fewer than 100,000 tracks, the database is automatically saved every five minutes when changes have been made. This helps protect cues, grids, analysis, playlists, and History rather than waiting until you exit to save the session.
• The database is written to a temporary file and swapped in only when complete. A crash during saving can no longer leave a partially written library.
• A damaged database detected at startup can be replaced with a backup. A failed save at exit is reported.
• Quitting during startup can no longer overwrite the library with an empty one.
• Track number and year sort numerically, placing “2” before “10.”
• List filters: fixed a BPM-only filter that rejected every track and a year-end-only filter that allowed every track through.
• M3U import now supports files with Windows line endings, a byte-order mark, very long lines, no final line break, relative paths, and non-ASCII paths. The .m3u8 extension is also accepted. Export writes a proper #EXTM3U header with durations and titles.
• Exporting M3U or CSV from the browser context menu exports the playlist you right-clicked, not the list currently displayed on screen.
• Remove dead files no longer removes streaming tracks or downloaded offline tracks.
• Adding files or folders adds only tracks that are not already in the target list.
• Energy and Mood columns are added to saved column layouts automatically.
• The tag editor writes back only the fields you edited.
• The track context menu includes a Lyrics submenu. Fixed cases where Toggle offline and Filter history by date were enabled for the wrong item.
• Cloud tracks that were only browsed and never used are no longer stored in the library file, making it noticeably smaller. The file also maintains a stable order between saves.
• Mac: Apple Music playlists are read in the background. Previously, a large playlist could freeze the window for minutes.

Storage and Backups

• Added a new Storage & Backup tab in Preferences.
• Storage location: move the entire DEX 4 data folder, including settings, database, skins, recordings, and karaoke stores, to a folder or drive of your choice. The move occurs at exit. Paths in the database are rewritten when the folder is found at a new location, such as a new drive letter or renamed share.
• Automatic backups of the Database, Settings, Shaders, and Overlays folders are created at exit. The previous run’s backup and the newest backups from the last day, week, and month are retained, up to a maximum of seven. Restore any of them from the Storage & Backup pane.

Mixer, Effects, and Microphone

• An automatic feedback suppressor on the On Air microphone reduces howling using up to 12 narrow notch filters. Filters are placed where feedback starts and released when no longer needed. The feature is enabled by default, with a checkbox on the Audio settings page.
• Fader, crossfader, trim, and EQ adjustments are click-free. See the Playback Engine section for details.
• Master VU meters are read without disturbing the audio thread.

Visuals and Video

• Add ShaderToy shaders by pasting code, the default method, or opening a local .frag, .glsl, or .txt file. The URL option remains available, but the ShaderToy API is rate-limited and often blocked.
• Clear error messages appear when a shader cannot be converted, instead of silently saving an empty shader.
• Mac: shaders run natively on Metal, with OpenGL retained as a fallback. The shader converter is universal, so Apple Silicon Macs no longer require Rosetta for shader conversion.
• Windows: the shader converter is now a current 64-bit build and is included in the installation again. ShaderToy conversion previously failed on most Windows installations.
• HD videos display the correct colors using BT.709. Partially updated video frames are no longer shown.
• Video recording captures complete frames on Mac and avoids stalling playback on Windows.
• Mac: a video output window on a second display is paced by that display.

Streaming Services

• TIDAL, SoundCloud, and KSR: login tokens are refreshed before they expire, preventing stalls when loading a deck. A refused refresh logs the service out cleanly instead of showing it as connected while every request fails. The refresh token can no longer be lost.
• TIDAL retains offline mode and downloaded tracks when the account loses its online session during a set.
• Fixed an application hang at exit after logging out of TIDAL and logging back in during the same session.
• The Preview player and Filler music player can play streaming tracks.

Audio Devices and Controllers

• Audio devices running at 48 kHz, or any rate other than 44.1 kHz, previously played everything 0.34% too fast, with a small glitch at every buffer. Playback is now exact, and device-rate conversion uses a high-quality filter on Mac CoreAudio and Windows ASIO.
• Windows ASIO: fixed microphone and timecode input running at the wrong speed on devices using sample rates other than 44.1 kHz.
• Controller: a Karaoke mode notification lights the skin button and pad.

Help Us Get DEX 4.1 Ready for Release

Your feedback helps us understand how DEX 4.1 performs with different music libraries, controllers, audio devices, and show setups. Download the beta using the buttons above, try the new features, and test the everyday tasks you rely on.

Please report any issues in the comments below. Include:

• Your Windows or macOS version and computer specifications.
• Your controller, audio interface, and any other relevant hardware.
• The steps that led to the issue, what you expected to happen, and what happened instead.
• A screenshot or short video when possible.

We also want to hear what is working well and which improvements are making a difference in your workflow. Thank you for helping us test the biggest DEX 4 update yet.