1. Introduction & Key Features

My project is a real-time AI voice assistant built on an ESP32. It listens to your questions, transcribes them via Deepgram, queries ChatGPT for up-to-date answers, then speaks back through an I²S amp—all while showing each step on a 16×2 LCD display. This makes it a hands-free, web-connected assistant without needing a PC.

• Live Web Data: Fetch news, weather, stock quotes, or any current event.

• Local Logging: Records every question (WAV).

• Clear UI: 16×2 LCD shows “Ready for Ans...”, “record start...”, “Speech to text”, and the actual text.

• Portable Power: Runs from any USB charger or power bank—no batteries to worry about.

2. Hardware Design & Assembly

All pin-out and wiring details can be found in the Schematic and Layout section. In brief, the core modules are:

• ESP32-WROOM-32: Main controller with Wi-Fi & I²S

• INMP441 I²S Mic: Digital microphone

• MAX98357A Amp: I²S audio output to speaker

• microSD Module: WAV recording & logs

• 16×2 I²C LCD: Status display

• Pushbutton: Record trigger

3. Project Demonstration Video

See it in action:
https://drive.google.com/file/d/1I4ZKJhOAJsSllxNhRCoU3JL7TKS1CD7C/view?usp=drivesdk

4. Software Workflow & Code

st=>start: Button Press  
rec=>operation: Record WAV to SD  
stt=>operation: Deepgram STT → Text  
gpt=>operation: ChatGPT API → Reply Text  
tts=>operation: OpenAI TTS → Audio  
play=>end: Play via I²S → LCD shows steps  
st->rec->stt->gpt->tts->play

1. Initialize Peripherals (I²S, SD, LCD, Wi-Fi)

2. Record Audio: Hold button → WAV saved to SD.

3. Transcribe: Upload WAV to Deepgram → receive plain text.

4. Chat & TTS: Send text to ChatGPT → get reply → send reply to OpenAI TTS.

5. Playback: Stream audio through MAX98357A → display messages.

6. Loop: Ready for next query.

Code Highlights:

• lib_audio_recording.ino – I²S + SD card WAV writer

• lib_audio_transcription.ino – HTTPS POST to Deepgram STT

• lib_OpenAI_Chat.ino – ChatGPT Completions API handler

• lib_audio_tts.ino – OpenAI TTS playback routine

• main.ino – Orchestrates button, LCD, and state machine

5. Reference Code

Find the complete source code and documentation on GitHub:
https://github.com/kaloprojects/KALO-ESP32-Voice-ChatGPT

6. Step-by-Step Build Tutorial

1. Solder headers to each module.

2. Wire modules per the PDF’s pin diagram.

3. Flash firmware via Arduino IDE

4. Insert a FAT32-formatted microSD card.

5. Power via USB charger—watch the LCD boot.

6. Hold the button, speak—watch and listen!

7. Lessons Learned & Pitfalls

• I²S Buffering: Tune buffer sizes to prevent underruns.

• Network Timeouts: Implement retries for STT and GPT calls.

• Power Stability: Use a reliable USB supply to avoid drops during Wi-Fi.

Conclusion
This project seamlessly integrates embedded audio I/O, cloud-based AI, and real-time web access into a user-friendly device—ideal for anyone looking to explore AI assistants on the go.