Portable TTS Server V2 v2.0.1 - Clickable launcher and download helper

Portable TTS Server V2 - v2.0.1

19 speech engines plus Whisper, voice cloning where supported, subtitles, audio editing, a Windows CLI, and an authenticated HTTP API. Kokoro and its offline assets are bundled; other engines install into the portable folder when selected.

Distro source: aivrar/portable-linux-in-a-box. This app originated as its child distro. The public V2 Linux runtime is rebuilt from clean Ubuntu with the speech dependencies included.

Download and start

1. Download Download-TTSServer.exe below and open it.
2. Choose a writable local folder with 35 GiB free. The helper downloads about 6.3 GiB, verifies SHA256 checksums, extracts the app, and launches it. Allow time for the download and first import.
3. For later starts, open Portable-TTS-Server-V2/TTSServer.exe. Startup progress and errors appear in a window with a WSL setup guide button.
4. In Server, spawn a Kokoro 82M worker on CPU or an available GPU. In Testing, choose a built-in voice and generate speech.

Host requirements: Windows 10/11 x64, working WSL2, and virtualization. WSL setup can require administrator approval and a restart. The downloader does not enable Windows features. An NVIDIA GPU is optional and needs its Windows driver. The EXEs are unsigned, so Windows may display an unknown-publisher prompt.

The app contains Python, .NET, WebView2, Linux dependencies, PyTorch/CUDA user-space libraries, FFmpeg, and Kokoro, including pronunciation assets for its nine language groups. No separate app runtime installation is needed. Other engines need internet to install; gated models require access, Edge uses an online service, and Whisper weights are optional downloads.

What's fixed in 2.0.1

• A clickable TTSServer.exe is in the main app folder again, with the TTS icon and visible first-start progress.
• Download-TTSServer.exe assembles the full portable release automatically. It is a small native Windows program with no separate .NET or VC runtime requirement.
• Startup checks for missing payload files and WSL readiness before importing the Linux image.
• Extraction validates checksums and archive paths, publishes only a completed app folder, and refuses to overwrite an existing installation.
• The README and wiki describe the engines, new download/start flow, and the Portable Linux in a Box lineage.

Manual or offline transfer

Download all four .zip.part001 through .zip.part004 files, portable-manifest.json, and both Extract-Portable-TTS helpers into one folder. Double-click the CMD helper. Extract-Portable-TTS.cmd -Download can also fetch missing parts. GitHub's automatic Source code downloads do not contain the runtime.

Download cache remains in .tts-download beneath your chosen folder. Remove that cache after verifying the app works. Existing app folders are never overwritten; choose another destination for a fresh copy.

Moving and preserving your installation

Run Stop-TTSServer.cmd before moving or copying the app. Once it confirms the disk is released, copy the entire folder, including wsl/ext4.vhdx. If WSL retains the disk handle, use Copy-TTSServer.ps1 to transfer without stopping unrelated WSL apps. Keep the original until the destination is verified. See the wiki for commands and instructions.

Verification and limits

The root EXE passed a native launch test from a folder with spaces: bundled WebView2, connected app, TTS icon, portable downloads, and graceful shutdown. Extraction checks passed for valid archives, damaged parts, unsafe paths, missing launchers, and preservation of existing destinations.

The actual v2.0.1 release parts also passed extraction through the native downloader, first-import startup from a new folder with spaces, and fresh CPU Kokoro generation: a 24 kHz, 5.66-second WAV. All 10 GitHub asset sizes and SHA256 digests matched. Anonymous downloads of the EXE, manifest and helpers matched the built files; all four archive URLs were reachable. Automatic approval review blocked an additional test that would execute the GitHub-downloaded EXE, without a detailed reason. The identical locally built EXEs passed the native tests described above.

The unchanged clean Linux image previously passed offline CPU synthesis for all nine Kokoro language groups, CPU and RTX 3060 API/CLI generation, and relocation with byte-for-byte preservation of saved audio. The existing regression suites passed 74 Python tests and 10 browser tests. Windows 10 build 19045 was the test host; Windows 11 and inference across all optional engines were not separately tested. The native same-folder duplicate-session scenario remains unverified; the backend duplicate guard was tested separately.

Source is MIT licensed; bundled components retain their own licenses, notices, inventories, and corresponding Ubuntu sources. Private development disks, voices, jobs, credentials, and browser profiles are excluded.

Portable TTS Server V2 - v2.0.0

Portable TTS Server V2 — v2.0.0

A portable Windows speech studio with its own WSL2 Linux distro, native desktop window, CLI, and authenticated HTTP API.

Download and start

1. Download Extract-Portable-TTS.cmd and Extract-Portable-TTS.ps1 into the same folder.
2. Run Extract-Portable-TTS.cmd -Download from a terminal. It downloads the manifest and four archive parts, verifies their SHA256 checksums, and extracts the complete app.
3. Run Portable-TTS-Server-V2\Start-TTSServer.cmd.
4. In Server, spawn a Kokoro worker on CPU or an available GPU; generate from Testing.

For manual downloading, get all four .zip.part001–.zip.part004 files, portable-manifest.json, and both extraction helpers, then double-click the CMD helper. The approximately 6.3 GiB download is split because GitHub limits individual release assets to under 2 GiB. The automatic Source code downloads do not contain the runtime.

Use Windows x64 with WSL2 and virtualization enabled. Allow at least 30 GiB free for extraction, first-run import, and working room, plus space for retained archive parts. An NVIDIA GPU is optional; it requires the Windows driver. The launcher is unsigned.

Included

• Offline-ready Kokoro, all built-in voices, English pronunciation model, Japanese dictionary, and Chinese pronunciation dependencies.
• Fresh Ubuntu 24.04.4 runtime with Linux Python, PyTorch/CUDA user-space libraries, FFmpeg, and audio dependencies.
• Windows Python, a self-contained .NET desktop launcher, and bundled WebView2 Fixed Version.
• New TTS icon, illustrated manual with eight real app screenshots, source code, license inventories, and corresponding Ubuntu package sources.

Other engines install into this copy's Linux disk when chosen. Those installations need internet; gated models require access. Edge uses an online service. Optional Whisper transcription weights are downloaded separately.

Moving and preserving your installation

Run Stop-TTSServer.cmd. Once it confirms the disk is released, copy the entire folder, including wsl/ext4.vhdx. If older WSL keeps the disk locked, run Copy-TTSServer.ps1 -Destination 'D:\Portable-TTS-Server-V2'; this exports the app's Linux state without stopping unrelated WSL apps. Keep the original until you verify the destination. See the included PORTABILITY.md for snapshot space and cleanup instructions.

Verification and limits

• 74 Python regression tests, 10 browser logic tests, and manual link/content checks passed.
• Fresh release-image import passed real CPU synthesis for all nine Kokoro language groups with Python network connections blocked. Real API/CLI speech passed on CPU and an RTX 3060.
• Transfer to a new folder preserved an existing job's audio byte-for-byte, allowed a new edit, and passed fresh GPU synthesis with correct discovery and paths. An unrelated WSL app stayed running. Incomplete transfers and nested destinations were refused.
• The bundled native browser displayed the connected app and TTS icon, and shut down successfully on Windows 10 build 19045. Final download-directory and duplicate-close changes are compile-verified and still need a native runtime retest.
• Optional engines and Windows 11 were not all separately tested. The release bundles app dependencies; Windows, WSL2, virtualization, and GPU drivers remain host requirements.

The public runtime was built from a fresh official Ubuntu image. Private development disks, personal voices, jobs, tokens, and browser profiles are excluded. Application source is MIT licensed; bundled components retain their own licenses and notices.

Final package integrity

All 1887 archived file hashes and all four part hashes were independently verified. The tested Linux image is unchanged. Source commit: 7ced9587472f15284a3a3633f0de8000bb2e59f2.

Joined archive SHA256:

f9a9b0699f6556d54f61349147d46cebe2508279ab98026758d1079afde69694