whisper.cpp runs OpenAI's speech recognition model in plain C++ with zero runtime allocations. It supports Apple Silicon, Vulkan, and NVIDIA GPUs, allowing faster-than-realtime transcription on devices from Raspberry Pi to Mac workstations. The base model requires roughly 388 MB of RAM, making it viable for embedded offline use.
It is a solid choice if you need local, dependency-free ASR without sending audio to the cloud. The main catch is that the CLI tool only accepts 16-bit WAV files, requiring a conversion step for other formats. Do not expect it to replace a full NLP pipeline; it strictly handles raw audio to text.
more: https://github.com/ggml-org/whisper.cpp
my other side: ivo.blogspot.com
Leave a Reply