Repository navigation
Releases
VoiceVault 0.2.1 — replay, pacing and control
VoiceVault 0.2.1 — replay, pacing and control
🔁 Replay any spoken reply
Every spoken reply in Converse now has a Replay button, and it keeps working after you switch chats or restart. Older replies offer Speak: it voices the reply now and keeps the audio for next time.
⏱️ Pacing
- Generate: a Pace setting (Tight to Dramatic) sets the silence between sentences.
- Exact control with tags, for every voice model:
[pause 0.8s],[slow]…[/slow],[fast]…[/fast]and[speed 0.7]…[/speed]. - Converse: each character can have a relaxed or dramatic pace between sentences.
🎚️ Voice Lab strength
- Voice change strength and Effects strength (0–150%) dial any preset up or down: "Elderly" at 60%, a softer robot, or an exaggerated ghost.
- Each effect also has its own Amount slider.
📦 Choose Drama 3's size
- Settings → Models can now build Drama 3 at Q2, Q3, Q4, Q5, Q6 or Q8, and keep the downloaded original so you can build more sizes later without downloading again.
- Smaller sizes use less memory but can lose quality, and the app says so before you build:
- Q2 and Q3 are labelled "lower quality" and need a confirmation.
- Every size uses the same proven conversion method; Q4 and Q6 are the ones tested on our PC.
- Details: docs/QUANTIZATION.md
🛠️ Fixed: NVIDIA setup
First-run setup failed on NVIDIA graphics cards because of a package version clash. Every card now gets the same PyTorch 2.9.1:
- NVIDIA: CUDA 12.8, or CUDA 12.6 if your driver is older than 570.
- AMD: ROCm 7.2.
- Everything else: the CPU build.
If an earlier setup left the wrong version, installing 0.2.1 replaces it automatically. Thanks to @DeeplyParallel for the report and diagnosis (#1).
🤖 For agents
vv_voice_shape takes strength and effects_strength; vv_speak takes pace.
Full list: CHANGELOG.md
Install or update
- Download
VoiceVault-Setup-0.2.1.exebelow and run it. It updates an existing install and keeps your library, voices, chats and settings. - Update notices: if you're on 0.2.0, VoiceVault can show this update in its banner and install it from inside the app.
- SmartScreen warning: Windows may warn because the installer isn't code-signed (certificates cost money every year and VoiceVault is free). Choose More info → Run anyway.
- Check your download (optional): in PowerShell run
Get-FileHash .\VoiceVault-Setup-0.2.1.exeand compare with:
SHA-256: b101e3fcc28d05f83ff3258ed078435a322d9df6e711097537d648c19bbc5918
Thank you
VoiceVault stands on the work of many open-source teams (CREDITS.md). If it's useful to you, you can support it on Patreon or Ko-fi. Every feature stays free for everyone.
This product uses the TMDB API but is not endorsed or certified by TMDB.
VoiceVault 0.2.0 — the Cast release
VoiceVault 0.2.0 — the Cast release
🎭 Cast: set a character up once, use them anywhere
The new Cast page keeps each character's portrait, full persona and voice, ready for any chat or scene.
- Craft with AI, lore-checked. For an established character, VoiceVault reads their Wikipedia article, and your chat model drafts the persona from it. Then a second pass must quote the exact sentence behind every statement, and VoiceVault confirms the quote really is in the source.
- Nothing unconfirmed goes in. Guesses ("likely", "may"), unconfirmed facts and spin-off media (novels, comics, games) stay out of the persona. They're listed under Open questions for you instead.
- Sources are shown with links, and every save keeps the previous version in History.
- Original characters. Describe an idea and get a full persona and voice description.
- Voice auditions. Maya1 designs three voices, and Whisper checks each take: garbled ones are flagged and the best match is marked. Drama 3 then performs the one you pick.
- Craft personas for all builds every favourite in one go.
- Add a source checks lore you paste from any page.
🙋 Your personas
Choose who you play in each chat or scene: a name, an avatar, how characters should address you, and notes. Characters respond to your persona, and your avatar shows next to your messages.
🔔 Update notifications
From this version on, VoiceVault tells you when a new release is out:
- The check: once a day it asks GitHub for the newest release. Only that question is sent, nothing about you.
- The download: the banner downloads the installer inside the app, then checks it against GitHub's SHA-256 fingerprint.
- Install now closes VoiceVault and runs the installer.
- Turning it off: Settings → General → Updates.
🤖 For agents
Seven new MCP tools (vv_cast_* and vv_persona_*). Ask your agent to "craft personas for my favourite characters" or "make me a persona". Like every VoiceVault tool, they can't delete anything.
Full list: CHANGELOG.md · Details: docs/CAST.md
Install or update
- Download
VoiceVault-Setup-0.2.0.exebelow and run it. It updates an existing install and keeps your library, voices, chats and settings. - Windows SmartScreen may warn because the installer isn't code-signed (certificates cost money every year and VoiceVault is free). Choose More info → Run anyway.
- Check your download (optional): in PowerShell run
Get-FileHash .\VoiceVault-Setup-0.2.0.exeand compare with:
SHA-256: f34326497271a13f590c9c888cbab64d6bdc4e5fb9b9a91ac891f92c6fc1889d
Requirements, first-run setup and model downloads are unchanged; see the README. Lore crafting needs a chat model (for example LM Studio).
Thank you
VoiceVault stands on the work of many open-source teams (CREDITS.md). Character lore comes from Wikipedia (CC BY-SA 4.0). If VoiceVault is useful to you, you can support it on Patreon or Ko-fi. Every feature stays free for everyone.
This product uses the TMDB API but is not endorsed or certified by TMDB.
VoiceVault 0.1.1
VoiceVault 0.1.1 — the trailer release
🎬 Watch the trailer
Watch on YouTube (7 min), or download VoiceVault_trailer.mp4 below with English captions (VoiceVault_trailer.en.srt).
Every voice, note and sound effect in it was made with VoiceVault:
- the narrator, characters and their emotions with Fish Audio Drama 3;
- voices designed from text with Maya1;
- the score with YuE2, with its vocals removed in Music Studio;
- the sound effects with Stable Audio 3.
All voices are synthetic.
What's fixed
Found while making that trailer with VoiceVault itself:
- Faster character chat with LM Studio. Reasoning models used to "think" silently for 5–20 seconds before the first word. VoiceVault now asks them to answer directly.
- Drama 3 handles low memory gracefully. Before loading, it checks there's enough free system RAM for the part of the model that doesn't fit on your graphics card, and tells you what to do if not. After an out-of-memory crash it now recovers on the next request instead of staying "not loaded".
- Voice Lab ageing sounds like age. Elderly, Very old and the Age slider add a natural shake, roughness and a slower pace, instead of just making the voice deeper.
Full list: CHANGELOG.md.
Install or update
- Download
VoiceVault-Setup-0.1.1.exebelow and run it. It updates an existing install and keeps your library, voices, chats and settings. - Windows SmartScreen may warn because the installer isn't code-signed (certificates cost money every year and VoiceVault is free). Choose More info → Run anyway.
- Check your download (optional): in PowerShell run
Get-FileHash .\VoiceVault-Setup-0.1.1.exeand compare with:
SHA-256: 0405a489019a71930211b5e5c70cf062b4d25b0adea8e3019e4676faf4746154
Requirements, first-run setup and model downloads are unchanged from 0.1.0; see the README.
Thank you
VoiceVault stands on the work of many open-source teams (CREDITS.md). If it's useful to you, you can support it on Patreon or Ko-fi. Every feature stays free for everyone.
This product uses the TMDB API but is not endorsed or certified by TMDB.
VoiceVault 0.1.0
VoiceVault 0.1.0 — first public release
A local-first voice and character studio for Windows. Free and open source, forever. No account, no subscription, no tracking: your audio, voices and chats stay on your PC.
What's inside
- Generate: text to speech in any voice. Clone a voice from a 10–30 s clip or design one from a description. Fish Audio Drama 3 by default, plus Kokoro, Maya1, OmniVoice and VibeVoice.
- Converse and Scene Creation: chat with characters (local LLM via LM Studio, Ollama or KoboldCPP, or a cloud API) and hear every reply. Long-term memory keeps chats coherent over thousands of turns.
- Voice Lab: change age, pitch, gender and speed, add effects, or turn your performance into another voice.
- Harvest: turn a recording into a clean voice clip: remove music, keep one speaker, restore quality.
- Music Studio: compose songs from lyrics and a style (YuE2, ACE-Step), split stems, analyse tempo and key, make sound effects.
- Discover, Collections and Library: browse and favourite shows, films, games and characters; every piece of audio in one searchable list.
- Models: one-click downloads from official sources, each with its licence and whether you may sell what you make.
- AI agents: MCP tools let a local model or Claude Desktop speak lines, reshape voices and compose music. Agents can't delete anything.
- Nine themes, built-in guides on every page, and a first-run setup wizard.
Install
- Download
VoiceVault-Setup-0.1.0.exebelow and run it. It installs for your Windows account only; no administrator rights needed. - Windows SmartScreen may say it "protected your PC", because the installer isn't code-signed (certificates cost money every year and VoiceVault is free). Choose More info → Run anyway.
- The first start sets up VoiceVault's private Python engine. It downloads several gigabytes (Python, PyTorch for your graphics card, the voice packages, and the Prisma database engine) and takes 10–40 minutes, once.
- Install FFmpeg if you don't have it (
winget install Gyan.FFmpeg), then get models in Settings → Models.
Check your download (optional): in PowerShell run Get-FileHash .\VoiceVault-Setup-0.1.0.exe and compare with:
SHA-256: b2b3c8d8cc913ec95d78d213160d3c7628336347c88a78d3fcdcfeada0773fdb
Requirements
- Windows 10 or 11, 64-bit; 16 GB RAM (32 GB or more recommended).
- A graphics card with 8 GB or more of VRAM for Drama 3; 16 GB or more recommended. CPU-only works for Kokoro, Voice Lab, transcription and chat, slowly.
- About 30 GB of disk for Drama 3, 100 GB or more for every voice and music model.
- Tested on an AMD Radeon RX 7900 XT (Vulkan and ROCm). NVIDIA and CPU-only are community-tested: please report how it goes.
Good to know
- Pre-release software. Expect rough edges; bug reports are very welcome.
- What you make is yours, as far as the models you use allow. Each engine shows whether commercial use is allowed. See MODEL_LICENSES.md.
- VoiceVault never ships model weights; it downloads them from their official sources after you accept each licence.
- You are responsible for what you create and must follow the law. See the terms of use.
Thank you
VoiceVault stands on the work of many open-source teams; see CREDITS.md. If VoiceVault is useful to you, you can support it on Patreon or Ko-fi. Every feature stays free for everyone.
This product uses the TMDB API but is not endorsed or certified by TMDB.