lovelyterry/SmartSpeaker

一个基于云端语音识别的智能控制设备，类似于天猫精灵，小爱同学。采用的芯片为stm32f407,wm8978,esp8266。

What it solves

This project implements a smart speaker device similar to Amazon Echo or Tmall Genie, but built using a microcontroller (MCU) and FreeRTOS instead of embedded Linux, demonstrating a lightweight approach to voice-controlled hardware.

How it works

The system uses an STM32F407VET6 as the main controller, a WM8978 for audio DA/ADC, and an ESP8266 for network connectivity. It employs a simple Voice Activity Detection (VAD) algorithm based on short-term zero-crossing rate and sound wave energy to automatically record audio. Once recorded, the audio file is sent to the Baidu Cloud speech recognition server for processing, and the device executes actions based on the returned recognition results.

Who it’s for

Developers and hobbyists interested in embedded systems, MCU-based voice recognition, and integrating cloud-based AI services into hardware.

Highlights

  • MCU-based architecture: Built on STM32F4 and FreeRTOS rather than a full Linux OS.
  • Cloud-integrated: Uses Baidu Cloud for speech-to-text processing.
  • VAD implementation: Includes a custom Voice Activity Detection algorithm for automatic recording.
  • Extensible: Designed to be easily expanded with new commands and actions, such as controlling LEDs or playing music from an SD card.

Related

  • Project
    Colin-XKL/FeedCraft

    FeedCraft is an open‑source Go service that acts as middleware for RSS/Atom feeds. It can fetch full articles, run AI‑powered summarisation, translation, and content cleaning via any OpenAI‑compatible LLM, and re‑publish the result as a new RSS feed. Users define **AtomCrafts** (single steps) and combine them into **FlowCrafts** (pipelines), then save them as **Recipes** that expose custom RSS URLs. Deployment is Docker‑compose‑based, with optional Redis caching and a headless‑browser service for JavaScript‑heavy sites. Licensed under GPL‑v3.

  • Project
    AIGAnimation/CAMDM

    A real-time character animation system using diffusion models to generate high-quality, diverse movements based on user control signals and style presets.

  • Project
    ACDSLab/MPPI-Generic

    A C++/CUDA header-only library for Model Predictive Path Integral (MPPI) control, enabling high-performance stochastic trajectory optimization on NVIDIA GPUs.

  • Project
    Oqura-ai/deepdoc

    deepdoc is a Python tool that turns a local folder of PDFs, docs, slides, images, and text files into a structured markdown report. It indexes the content in Qdrant, uses an LLM to plan the report outline, and runs parallel research agents that query the index, filter results with TypeSafe’s Jev model, and synthesize each section with a writing LLM. All processing stays on‑device, and the pipeline is highly configurable via `configuration.py`.