Skip to content

audio.cpp Windows portable v0.4.2

Latest
Latest

Choose a tag to compare

View all tags
@kigner kigner released this 30 Jul 12:36
v0.4.2-windows-prebuilt
35dd208

audio.cpp Windows portable v0.4.2

This release combines the upstream audio.cpp 0.4, 0.4.1, and 0.4.2 changes
with the local Windows WebUI extensions in one stable portable update.

Highlights

  • Added GLM-TTS end-to-end support:
    • native CPU/CUDA runtime integration
    • zero-shot Chinese and English voice cloning
    • WebUI catalog, parameters, required-file checks, and model status
    • model-manager download and conversion support
  • Updated the WebUI's model-spec handling so supported named GGUF files and
    model directories are detected consistently.
  • Changed CPU thread defaults to use physical cores, while keeping explicit
    user overrides available.
  • Added upstream runtime improvements including reusable framework modules,
    Hugging Face-compatible sampling, faster GGUF loading on Windows, and
    reusable Chinese text normalization.
  • Fixed the Qwen3-TTS speech decoder graph cache so a graph is reused only
    when its exact frame count matches the request.
  • Included the upstream CUDA random-generator probe fix, WebUI model-download
    fix, and Windows prebuilt-package parsing fix.

Windows builds

  • CPU core: AVX2
  • CUDA core architectures:
    61-real;75-real;86-real;89-real;120a-real;90-virtual
  • Updated executables: audiocpp_cli, audiocpp_server, and audiocpp_gguf
  • Python environment changes: none
  • CUDA runtime DLL changes: none
  • Updater version: 1.1.1

Updating

Run update.bat from an existing Windows portable installation. The stable
channel updates directly to v0.4.2; no intermediate v0.4.0 or v0.4.1 package
is required.

  • Supported source versions: 0.2.0 and later
  • Publicly released source versions verified: 0.2.1, 0.2.2, and 0.3.0
  • Development-version metadata accepted: 0.4.0 and 0.4.1
  • Updated components: app, core-cpu, and core-cuda

The updater preserves downloaded models, custom voices, generated output, API
keys, logs, UI language preferences, and SpeakType configuration. If
validation fails, managed files are restored from _update\backup\.

Validation

  • CPU and CUDA builds were manually tested before packaging.
  • WebUI and model tests passed, including the Qwen3-TTS decoder fix.
  • The signed final assets are validated against every supported source-version
    case before publication.
  • Release manifests, checksums, component indexes, and remote assets are
    verified with the bundled minisign public key.
Assets 11
Loading