llama.cpp Release b11675 Adds Multimodal Audio Streaming Support
Release b11675 of llama.cpp introduces `mtmd_audio_stream` along with integrated support for the `voxtral-realtime` model. This addition enhances the library's multimodal processing capabilities for live audio applications. This update makes it easier for developers to build responsive voice agents and low-latency real-time transcription tools locally. By expanding multimodal support, llama.cpp continues to bridge the gap between heavy cloud infrastructure and localized open-source AI execution. The update features changes under the `mtmd` module via pull request #30629, implementing the underlying audio stream capability. Pre-compiled binaries and cross-platform compilation packages are provided across major operating systems including macOS, Linux, Android, and Windows.
## BACKGROUND
Llama.cpp is an open-source software library that runs inference on large language models locally and serves as a foundational backbone for local AI tools. Voxtral Realtime is a speech-transcription model designed for live transcription and voice agents with ultra-low latency.