LEARNING NOTES
English translations are generated with an LLM and published only after human review.
How one ESP32-S3 prototype moved from a complete but sluggish record–transcribe–respond–synthesize pipeline to one real-time voice session—and why the hard part was coordinating audio, networking, display, and playback on constrained hardware.
A postmortem of severe ESP32-S3 recording undersampling: from a Base64 buffer error to identifying low-speed ST7789 redraws that blocked I²S, then validating the fix with a dedicated FreeRTOS capture task and rate-limited partial refreshes.
Documents the tested ESP32-S3 and INMP441 pipeline, from press-and-hold recording, on-screen volume feedback, and WAV upload to server-side audio processing, whisper.cpp transcription, and web display.
The local voice platform originally called TTS again whenever the same text was sent, creating a duplicate MP3. Starting with that real duplicate-file problem, this article explains why audio assets and playback jobs should be separate, and how cache keys, history playback, resending, and legacy database migration fit together.
I moved the ESP32-S3 from playing audio embedded in firmware to speaking text entered dynamically in a browser. This article documents the working path from the web UI through local TTS and a task queue to the ESP32, MAX98357A, and speaker—and clearly marks the voice-input work that remains.
A hands-on ESP32-S3 voice playback build: adapt a two-pin speaker connector to the MAX98357A screw terminal, complete the five-wire I²S connection, and verify the audio output with three test tones and five web-generated voice clips.
To practice soldering while ending up with something usable, I chose a YCA2109 electronic dice practice board. This article records how I started with a bare board and a bag of components, spent about five or six hours assembling, soldering, and troubleshooting it, and finally got the LEDs to light briefly.
A freshly soldered YCA2109 electronic dice had all LEDs unlit when powered; the voltage after connecting the battery holder to the circuit board was only about 2.2V, and the two chips became noticeably hot after about 10 minutes. This article documents how I worked through the power supply, button, multimeter range, and input diode step by step, eventually finding that the two DIP chips were in the wrong orientation; after flipping them, the electronic dice resumed operation and could randomly stop on different numbers.
When I first connected the GMT130-V1.0, I could only follow the wiring table. This article groups VCC, GND, BLK, RES, SCK, SDA, and DC into four categories, and walks along the complete display chain to explain how they turn ESP32 code into a visible image, and how to safely re-plan the wiring once you understand it.
