Question 1

How to fix unnatural hard-cut silence at [pause] markers in VoiceStudio voice cloning?

Accepted Answer

Pauses sound gated because `[pause]` inserts exact digital zeros while cloned speech carries a natural noise floor (~−53 dBFS). Fix: enable the room-tone fill added in PR #3615 via Settings → Pronunciation dictionary → Pauses → **Fill pauses with room tone**, or set the environment variable `OMNIVOICE_PAUSE_FILL=room_tone`. The filler noise is spectrally shaped from the take's own quietest frames at matching RMS level, uses a fixed seed for reproducible renders, and crossfades 40 ms into the surrounding lines. It applies to `[pause]` markers in Clone and Speak plus audiobook/story line and paragraph gaps. Digital silence remains the default, and silence is still used when no usable quiet floor is detected. Note: short zero-gap runs at long-text chunk joins are not covered by this fix and should be reported separately.

Question 2

Why won't CSM voice cloning work with MLX-Audio on Apple Silicon, and how to fix it?

Accepted Answer

Installing `mlx-audio` only installs the backend wrapper — curated submodels like CSM are downloaded/cached lazily on first load, so the model can appear unselectable until it's been loaded once. Fix: prewarm the model manually from your VoiceStudio backend environment: `VoiceStudio/project/.venv/bin/python -c "from mlx_audio.tts.utils import load_model; load_model('mlx-community/csm-1b-8bit')"`. After the download/cache completes, return to VoiceStudio and CSM (voice cloning) will select and work normally.

Question 3

How to install VoiceStudio TTS engine venvs to a custom location instead of the C drive?

Accepted Answer

Set the `OMNIVOICE_ENGINES_DIR` environment variable to your desired path (e.g., a folder on another drive). This moves the engine virtual environments and the audio.cpp runtime to that location. The default C drive location remains unchanged unless you set this variable. Available since PR #3024 on main, shipping in the next release.

Question 4

Why are emojis and supplementary Unicode characters split into surrogate halves or replacement characters in Stories spelling markup?

Accepted Answer

The JavaScript spelling helper used `split('')`, which iterates UTF-16 code units and breaks supplementary Unicode characters into isolated surrogate halves. Replace code-unit splitting with code-point iteration, for example `Array.from(text)` or `[...text]`, instead of `text.split('')`. This preserves emojis and supplementary letters as single characters and matches the Python parser behavior. The fix is included in the PR "fix(stories): preserve Unicode code points in spelling markup".

Question 5

How to fix VoiceStudio backend startup error "'NoneType' object has no attribute 'splitlines'" on Windows?

Accepted Answer

This crash happens in VoiceStudio 0.5.7 on Windows when the GPU-name probe reads PowerShell output containing characters invalid in the system locale encoding. Python's reader thread fails silently, leaving the output as None, and `result.stdout.splitlines()` in `backend/api/routers/system.py` throws `AttributeError`, killing the backend during the 'api_routes' startup phase (exit code 1, health endpoint returns 503). Fix: update VoiceStudio to the release after 0.5.7 — the fix (PR #3615) makes every text-mode tool call in the backend decode safely so the GPU probe can no longer block startup. No manual workaround is required; just upgrade and restart the app.

VoiceStudio

Core Features

What It Can't Do

Use Cases

Tags

Getting Started

Similar Projects