← PODLAB / The appsUSER MANUAL · BETA · 13 OCTOBER 2026
POTTS-VOX
Post Office Text To Speech - VOX
Local temporary dialogue and performance voice conversion.
Before you begin
Use a clean, dry 5–15 second single-speaker voice reference, with permission. Avoid music, echo and background voices. Put representative pronunciation near the start. Models download on first use and are cached for subsequent local processing.
Convert a performance
Select performance conversion. Choose your acted line as the performance recording and the target speaker recording as the voice reference. Choose an output folder and filename, confirm voice permission, then generate. Audition against picture: timing, pronunciation and emotion may change.
Generate typed dialogue
Select Reference-guided dialogue (English) and enter a short English line. Choose a reference spoken with the desired accent. Start with Reference guidance 0.65, expression 0.50 and variation 0.80. Confirm permission and generate. Try several short takes; accent retention is not guaranteed.
Understand the controls
Guidance controls how strongly generation follows the reference; it is not a dedicated accent control. Expression changes exaggeration. Variation changes sampling temperature. Adjust one control at a time. Use last seed reuses the previous seed; hardware and model changes can still affect repeatability.
Export and manage memory
48 kHz / 24-bit WAV is the default editing format. Model-native 24-bit WAV is also available; upsampling does not add detail. Play Last Take previews the result; Show Last File reveals it. Release Engine frees model memory. Disable Keep engine loaded to release it after each take. Cancel stops the current worker.
If a take fails
Check input files, a writable output folder and enough free memory. Confirm the first model download completed. Try Release Engine and a shorter line. Logs normally live at ~/Library/Logs/POTTS-VOX/potts-vox.log. Keep original recordings and review every generated take.
