The latest release of raullenchai/esp32-thinks is v0.2.11. This repository is a project that builds a personal voice assistant on the ESP32-S3. A user says the wake word "Hi, ESP" or presses a knob to ask a question, and receives spoken answers in English, Chinese, or Korean. However, the provided facts list nothing that changed in v0.2.11. So the specific changes in this release cannot be described based on confirmed information alone.
The core of this project is connecting OpenAI-compatible servers separately for speech recognition (STT), conversation, and text-to-speech (TTS), so that providers are combined per stage. Instead of putting all speech processing on a single service, you can choose a different compatible server for each task. A configuration that streams responses sentence by sentence, handles interruption during conversation, and covers reconnection is also presented in the README summary. The repository also includes an optional openmuse agent that provides location, time, weather, and persistent memory features. The underlying SDK is muse-gadget-sdk, and the license is MIT.
The README is recorded as containing per-device measurements and limitations, setup instructions, and a description of a patch applied to the underlying SDK. This is a clue to examine which devices and configurations the results targeted before use. However, the per-device measurements do not guarantee the performance of every ESP32-S3 configuration, and the provided facts include no standardized comparative benchmarks either. Since no separate release change log is given either, the appropriate scope for judging v0.2.11 is the speech processing structure the repository describes and the extent of its verification, rather than new features.