Built in public.

Faster voices, a better noise remover, new stem splits

• Create is faster. Voice conversion now runs on your Mac's graphics chip, so a 3 minute vocal converts in well under a minute on recent Macs. Files come out at the voice's own sample rate, and long songs use far less memory.
• Auto pitch. When a female voice gets a low vocal, or a male voice a high one, Xport moves it an octave for you, tells you, and lets you undo it.
• Mix/Master. A new Master Polish sound: warmer, wider and louder, and it lands on the loudness it shows without going past a safe peak. Before and after compares your original at the same volume. Auto Mix keeps your 808s.
• Split Stems add-ons (free): Drum kit split puts kick, 808, snare, toms, hi-hats and cymbals on separate tracks. Lead vs backing vocals splits the lead from harmonies, doubles and ad-libs. Nothing downloads until you click, and they work offline after that.
• Noise Remover now uses a real neural model (DeepFilterNet3), tuned to protect singing. Stereo files stay stereo.
• Library: Undo after deleting a voice, and drag voices into your own order.
• Fixes: Audio Converter, Trimmer, the stem add-ons and Train Clone no longer fail with "ffmpeg exit -9". Train Clone names the real problem when your audio can't be read. Text to Speech explains when a reference clip is too short. Quitting the app also stops its engine. The engine only answers the Xport app itself.
• Mac only. Existing Mac installs update automatically when you quit the app. Windows stays on alpha.41 until the Windows alpha.44 build is done.

Listen Live: key and BPM in real time

• Listen Live (free, part of Key & BPM Finder): pick a microphone or audio input under "Listening on" and the key and BPM show up live. The reading locks in after about 8 seconds.
• The mic is on only while Listen Live runs. Short audio windows are written briefly to a temporary file on your computer, analyzed there, and deleted. Nothing is uploaded or kept. The Privacy Policy now has a section on the microphone and live listening.
• Voice Blender (in Library): blends two voice models of the same RVC generation (v1 with v1, v2 with v2) into a new voice. If a pair cannot blend, the app tells you why. The compatibility check is free. Blending is Pro after the 7-day trial.
• Studio Voice (Pro after the 7-day trial): turns a quick Instant Voice into a trained voice model. The training runs on your Mac.
• Fixes: the V1 option in Train Clone's advanced settings is turned off, since only v2 training files ship and V1 always failed partway through. A damaged voice file no longer crashes the Voice Blender check. In-app legal pages no longer show stray asterisks.
• Mac only. Existing Mac installs update automatically when you quit the app. Windows stays on alpha.41 until the Windows alpha.42 build is done.

No more false "locked" banner

• AI voice conversion and Text to Speech now show as available when their packs are installed. Before, they read as locked if only the training pack was missing.
• Includes all the alpha.40 fixes, including license key activation.
• This is the current Windows build.
• Existing installs update automatically.

License activation fixed

• Activating a license key no longer fails with an error.
• New: activate your license from the profile menu.
• Installing the Voice Modeling Pack no longer fails at the training step on a fresh Mac.
• Key & BPM analysis is more stable.
• Existing alpha.39 installs update automatically.

Signed and notarized for Mac

• The Mac build is signed and notarized by Apple, so it opens without a security warning.
• The Mac build is for Apple Silicon only.
• Tools in this build: voice conversion, Train Clone, Text to Speech, Split Stems, Mix & Master, Key & BPM Finder, and more. They run on your computer.
• To install on Mac, open the .dmg and drag Xport Studio to your Applications folder.

Clean stable baseline before v2.0

Train Clone is now visibly marked "Coming Soon · v2.0" with a clear banner pointing users to the Library tab to import any RVC v2 model. Every user-facing surface in the app either works or is explicitly labeled as not-yet-available, no more clicking something that crashes silently after hours of CPU.

Stem Split, real fix this time

Resolved the torchvision PYZ bundling issue that broke Stem Split on packaged builds (UVR-MDX-NET, BS-Roformer, MDX23C, Mel-Band Roformer, Demucs all working again).

Engine stays online during long jobs

Fixed the heartbeat timeout that caused the engine to falsely report "offline" during long Train Clone or Mix & Master runs while work was still progressing in the background.

Hotfix: Train Clone + Stem Split

Patch on top of v1.1.13 fixing two regressions in the first stable baseline build.

First clean stable baseline

Zero-shot voice conversion explicitly marked "Coming Soon" instead of producing cryptic errors. Every UI surface in the app either works or is honestly labeled, the start of the beta program's quality bar.