From d55fae6335f66b3ca7ea567c3d73ba50a86a0155 Mon Sep 17 00:00:00 2001 From: "Danilo M." Date: Fri, 25 Sep 2026 19:12:09 +0200 Subject: whisper.cpp: add version 1.9.4 Worked ahead of SBo, which is still at 1.6.2. Upstream moved to ggml-org/whisper.cpp; HOMEPAGE and DOWNLOAD follow it. OpenVINO is required and always enabled. openvino installs under /usr/share/openvino, off the default linker path, so cmake is pointed at its config dir and its library dir goes into the install rpath, which lets the binaries run without sourcing setupvars.sh. OpenBLAS stays autodetected. Built with static libs and shipped as binaries only: the bundled ggml would otherwise install libggml*.so and ggml headers that clash with llama.cpp-vulkan, the same approach as stable-diffusion.cpp-vulkan. Test-built on -current with openvino as a local dep, lint clean. That needs OpenCL-CLHPP 2026.05.29, which the buildsystem gets from a slackrepo hint; SBo still ships 2024.05.08. Adds the nvchecker stanza and README row. Co-Authored-By: Claude Opus 5.5 --- whisper.cpp/README | 16 ++++++++++++++++ 1 file changed, 16 insertions(+) create mode 100644 whisper.cpp/README (limited to 'whisper.cpp/README') diff --git a/whisper.cpp/README b/whisper.cpp/README new file mode 100644 index 0000000..6daff42 --- /dev/null +++ b/whisper.cpp/README @@ -0,0 +1,16 @@ +High-performance inference of OpenAI's Whisper automatic speech +recognition (ASR) model, plus NVIDIA Parakeet models. + + - Plain C/C++ implementation without dependencies + - AVX intrinsics support for x86 architectures + - Mixed F16 / F32 precision + - Integer quantization support + - OpenVINO encoder support + +Installs whisper-cli, whisper-server, whisper-bench, whisper-quantize, +whisper-vad-speech-segments, parakeet-cli and parakeet-quantize. + +OpenVINO support is built in. The encoder must be converted to OpenVINO +IR format first, see "OpenVINO Support" in the bundled README.md. + +OpenBLAS is autodetected: if installed, it is used for the BLAS backend. -- cgit v1.2.3