# Hermes open-source firmware turns ESP32 into push-to-talk voice assistant for Windows PCs

> **Key takeaway:** Hermes is an MIT-licensed project that pairs an ESP32 with a Windows PC to create a low-cost, push-to-talk voice assistant using cloud-based speech processing via Groq or NVIDIA.

**Published:** 2026-10-29T21:05:47+00:00 
**Source:** Adafruit Electronics 
**Category:** open-hardware 
**Canonical URL:** https://fosswire.org/news/hermes-open-source-firmware-turns-esp32-into-push-to-talk-vo.html 

## Summary
Developer ROHITCRAFTSYT has released Hermes, an open-source voice assistant project that transforms an ESP32 microcontroller into a push-to-talk audio capture device paired with a Windows PC. The firmware, published under the MIT license, captures audio through an INMP441 digital microphone when a physical button is held, then transmits the clip over WiFi to a Python-based PC application for speech-to-text transcription and command processing using Groq's cloud API, with NVIDIA as an alternative backend. The PC responds with synthesized speech played either through the computer's speakers or an optional I²S amplifier and speaker connected to the ESP32. Build costs range from approximately ₹1,500 to ₹3,000 depending on component choices.

The project provides complete firmware source, wiring documentation, setup instructions, and a schematic for the ESP32-S3 variant, though the primary firmware targets the standard ESP32. The PC-side software requires Python 3.11 or newer. Voice commands can launch applications, open websites, and control smart home devices. Notable limitations include the absence of on-device wake word detection and the requirement for internet connectivity, as all voice processing occurs remotely rather than locally on the device.

## Hardware architecture and firmware design
- **MCU platform:** Primary firmware targets the standard ESP32; repository includes schematic for ESP32-S3 variant, both using Espressif's dual-core Xtensa LX6 architecture.
- **Audio I/O:** INMP441 MEMS digital microphone provides I²S audio input; optional MAX98357A I²S DAC/amplifier drives external 8Ω speaker for local audio output.
- **Connectivity:** WiFi used to transmit captured audio clips to PC; no Bluetooth or wired audio transport implemented.
- **PC software stack:** Python 3.11+ application handles transcription and inference via Groq API with NVIDIA as alternative provider; no local LLM or offline processing path.
- **Interaction model:** Push-to-talk button enables manual voice capture control; continuous listening or wake word detection not implemented, reducing standby power but requiring physical interaction.

## Why it matters
Hermes offers a fully documented, low-cost entry point for developers and hobbyists to experiment with voice interface hardware without proprietary smart speaker ecosystems, with total build cost under $40. The open-source release of firmware, schematics, and wiring diagrams aligns with Adafruit's community-driven hardware culture and provides a reference implementation for ESP32-based audio applications. However, the dependency on cloud APIs for transcription and the lack of offline processing or wake word detection limit its utility for privacy-conscious deployments and always-on home automation scenarios, positioning it as an educational and prototyping platform rather than a production alternative to commercial voice assistants.