llama.cpp Release b11781 Adds New POST /tts Endpoint for Text-to-Speech
Release b11781 of llama.cpp introduces a new POST /tts endpoint designed to support Text-to-Speech models. It includes audio output streaming capabilities and allows users to upload speaker reference samples via form-data. This update significantly expands the multi-modal capabilities of the llama.cpp server, making it much easier for developers to deploy local audio and language pipelines. It moves the framework beyond standard text generation into rich speech synthesis use cases. The implementation adds server routing and schemas specifically tailored for Text-to-Speech, while also fixing multipart form field handling so that parameters like seeds and boolean streaming flags are properly coerced from strings.
## BACKGROUND
llama.cpp is a widely used open-source library that enables efficient inference for large language models on consumer hardware. Built on top of the GGML tensor library, it provides both command-line tools and a local server backend that serves as the foundation for many popular AI applications.