Matrix Voice WebSockets Streamer

This repository allows you to build an offline ASR module based on WebSockets and MQTT protocols.

Note that this code was inspired by Matrix-Voice-ESP32-MQTT-Audio-Streamer.

Here are key differences in comparison with the original version:

• Project structure: the code was decomposed into logical blocks to increase readability and maintainability.
• MQTT: a sync client, which was initially used for audio streaming, was completely removed due instability (bugs that put ESP32 into zombie mode).
• Sockets: instead of a sync MQTT there was added a much more stable WebSockets client for voice streaming.
• Audio playback feature was completely removed due to instability and bad output quality.
• Settings: this version doesn't have such amount of customizations and doesn't persist configs in a SPIFFS.
• Everloop: there was added a new OTA animation and optimized the existing circular reaction on wake word detection event.

Setup

• Setup RPi and laptop/PC for the first flashing as described here.
• Import a source code into VSCode or CLion as a PlatformIO project.
• Adjust platformio.ini according to comments.
• Connect Matrix Voice to RPi.
• Run ./deploy [RPi_IP].

Voice Streaming

This code is configured to work with Vosk ASR out of the box. However, if you want to stream voice data to your own ASR engine, make sure your WS server can accept 1024-bytes chunks in a 16k sample rate.