whisper.cpp-1.9.4
automatic speech recognition (openai whisper model)
Project homepage: https://github.com/ggml-org/whisper.cpp
Description
High-performance inference of OpenAI's Whisper automatic speech
recognition (ASR) model. Features include:
- Plain C/C++ implementation depending only on GGML
- AVX intrinsics support for amd64
- VSX intrinsics support for POWER architectures
- Mixed F16 / F32 precision
- Integer quantization support
- Zero memory allocations at runtime
- Vulkan support
- Support for CPU-only inference
- C-style API
- Voice Activity Detection (VAD)
Package information
- Ports path
- audio/whisper.cpp
- Package architecture
- amd64
- Maintainer
- The OpenBSD ports mailing-list <ports@openbsd.org>
- Categories
- audio
- Available flavors
- None listed
- Only for architectures
- aarch64, amd64, arm, i386, mips64, mips64el, powerpc, powerpc64, riscv64, sparc64, alpha, hppa
These are ports metadata. Binary availability depends on the release, architecture and mirror. Build and test dependencies are not an installation checklist.
Direct dependencies
Library
Build
- devel/cmake/core STEM->=4
- devel/ninja
- devel/dwz
Test
OpenBSD README
+-------------------------------------------------------------------------------
| Running whisper.cpp on OpenBSD
+-------------------------------------------------------------------------------
whisper.cpp requires use of an AI model for operation. This package
bundles a base model for transcription that was trained on English
source material (ggml-base.en).
Other models provide higher accuracy at the expense of increased use
of resources, and there's a choice of multilingual or English-only
source material (for the smaller models, English-only tends to give
better results on English input - for medium/large, multilingual
may work better - test with your typical source material).
A download script, whisper.cpp-download-ggml-model, is provided.
To list available downloads:
$ whisper.cpp-download-ggml-model
To fetch the multilingual medium model:
$ whisper.cpp-download-ggml-model medium
Files fetched in this way can be used by passing the -m flag:
$ whisper-cli -m ggml-base.bin -t {threads} -f {sample}.wav
whisper-cli defaults to English; specify a language (fr, cy, eu, ...)
or "auto" with the -l flag. See https://whisper-api.com/docs/languages/