Nicheloom

Market intelligence for builders — see what's gaining traction before it's crowded.

hey-donut

A custom "Hey Donut" wake word for Home Assistant — 62 KB, runs on-device, zero false accepts per hour. Includes the full training pipeline so you can teach your assistant any name. CC0.

Details

External ID
1366248710
Source
GITHUB
Company
—
Product
hey-donut
Website domain
github.com
Launched
Sept. 11, 2026
Cohort
—
Upvotes
7
Upvotes percentile
0.05976172175249808
Tags
esphome, home-assistant, microwakeword, on-device, privacy, tensorflow-lite, voice-assistant, wake-word
Fetched at
Sept. 15, 2026, 1:03 a.m.
Updated at
Sept. 15, 2026, 1:03 a.m.

Enrichment

Theme
sleep, wake, and clock utilities
Vertical
Horizontal
Function
Model & infra
Audience
Developer
AI stance
AI-native
Project type
Hobby / open-source project
Normalized one-liner
on-device custom wake word engine for home assistant
Manually corrected
False

Could you build this?

Partial While wrapping the wake word into Home Assistant via openWakeWord or microWakeWord is straightforward, curating synthetic audio datasets and tuning false-accept rates to near-zero requires specialized ML audio pipeline work.

What it would actually take: Building this requires a wake-word training pipeline using PyTorch/TensorFlow with synthetic TTS generation (e.g., Piper), data augmentation with ambient background noise datasets (MUSAN, AudioSet), and model quantization to TFLite Micro for on-device edge inference. Expertise in acoustic modeling, phoneme alignment, and threshold calibration for zero false-positive tolerance is necessary.

Competitors

Other products that read as similar to this one — 34 launches clear the similarity bar, closest 8 shown.

Attention rank: #34 of 35 (itself plus its competitors, highest first — normalized so YC and Product Hunt are compared fairly).

Launched 251 days after the earliest competitor.

Other launches for this product

Same idea, different domain

Nobody's really built a model & infra tool for Fintech yet.