Technical Articles
Online Chinese Wake Word Training
Where to create and test a custom Chinese wake word model.
Online English Wake Word Training
Create and test a custom English wake word model.
Online Japanese Wake Word Training
Create a Japanese wake word model and deploy it offline.
Online French Wake Word Training
Train and test a custom French wake word.
Online German Wake Word Training
Test a custom German wake phrase before deployment.
Online Spanish Wake Word Training
Build an offline Spanish wake word model.
Online Hindi Wake Word Training
Create and validate a custom Hindi wake word.
Online Chinese KWS Training
Create single-keyword and multi-keyword Chinese command models.
Online English KWS Training
Train English keyword models for fixed voice commands.
Online Japanese KWS Training
Train Japanese keyword models for fixed commands.
Online French KWS Training
Create French KWS models for offline fixed commands.
Online German KWS Training
Create German offline command models.
Online Spanish KWS Training
Build Spanish keyword models for fixed voice control.
Online Hindi KWS Training
Create Hindi fixed-command keyword models.
KWS for Voice Control
Detect fixed commands locally and map them directly to device or app actions.
KWS in Voice Interaction: Wake First, Then Start ASR
Use Javis, Jarvis or other wake aliases to start ASR and the dialogue flow.
Wake Words in Voice Interaction
How a wake word starts listening and works with KWS, ASR, intent parsing and speaker verification.
Wake Words for Voice Control
From choosing a product name to triggering a reliable offline device action.
What Is a Wake Word?
How wake words differ from commands and ASR, and how to choose a reliable name for a voice product.
What Is openWakeWord?
Open-source self-training, online training options and the checks to make before deployment.
Voice Control vs. Voice Interaction
Understand the difference between fixed actions, KWS, ASR, intent parsing and dialogue systems.
How to Choose a Voice-Control Solution
Compare KWS, offline ASR, cloud ASR, voice modules and custom model platforms.
Offline Voice Control for IoT Products
Choose between KWS, ASR, voice modules and custom model delivery for an IoT product.
Custom Wake Word for Existing Hardware
Add offline wake words or fixed commands without replacing the main hardware.
How to Turn Your Product Name into a Custom Wake Word
Evaluate a product name, test it with real voices, and deploy it as an offline wake word.
KWS Advantages and Disadvantages vs. ASR
Compare keyword spotting with speech recognition, speaker verification and audio-feature matching.
What Is ONNX? A Practical Guide for Voice-Control Models
What ONNX stores, where it runs, what it does not solve, and how it fits into an offline voice-control product.
Offline Voice Control with Custom Wake Words and Multi-Keyword ONNX Models
Build a two-stage offline voice-control system: wake word first, fixed commands second, with R1 false-trigger optimization.
What Is a Voice-Control Model Solution?
From microphone input and model choice to runtime, anti-trigger logic, product actions and real-device validation.
KWS, ASR or Speaker Verification?
Choose the right voice model by asking whether the product needs a known phrase, open speech or speaker identity.
How Do We Make Multilingual Wake Words Reliable?
From language-specific data and TTS to real recordings, R0/R1 evaluation, and delivery gates across seven languages.
Offline Voice Control with ONNX/TFLite Models — From Demo to Deployment
A working ESP32-S3 demo (custom wake word + spoken commands, two INT8 models) and every step to reproduce it: run the free demo model, train free, test free, then decide what to optimize.
Same Data, Same Recipe, 5× Different: Training Variance in Wake Word Models
Production data across 5 languages: why two retrains of the same model differ by up to 5× in false triggers, and why every retrain now ships 5 candidates for you to pick from.
同一批数据训练，误触发差 5 倍（中文版）
Chinese version of the training-variance write-up: seed lottery in small KWS models and multi-candidate selection.
What Is a KWS Model? Offline Voice Control with ONNX
A practical introduction to wake words, fixed voice commands, local inference, and when KWS is a better fit than ASR.
Train an ONNX Wake Word Model Online in 2026
Enter a custom phrase, generate a portable model, and run wake-word detection locally on multiple platforms.
How to Train a Custom Wake Word: Self-Training vs Online Platforms
Compare setup, data, languages, testing and deployment for openWakeWord, Picovoice Console and Voicute.
Train an ONNX Keyword Spotting Model Online in 2026
Put several fixed voice commands in one compact KWS model and connect them directly to application actions.
Why openWakeWord Recall Varies by Wake Word
How a frozen speech-embedding backbone creates phrase-dependent separability, per-word tuning and “word picking.”
openWakeWord Not Detecting? Fix Low Recall Step by Step
Check model loading, 16 kHz audio, framing, scores, thresholds, VAD and training coverage before retraining.
Picovoice Free Trial Ended? Three Offline Alternatives
Train openWakeWord yourself, generate a portable ONNX/TFLite model, or stay with a commercial SDK.
Replace Picovoice on Python with an Offline ONNX Model
Run a custom wake word on Linux, Windows, or macOS without a runtime AccessKey.
Offline Keyword Spotting on Android
Detect fixed voice commands with ONNX Runtime and map them to app actions without cloud ASR.
Keyword Spotting in the Browser
Capture the microphone, run ONNX locally, and control the page without an inference backend.
Offline Voice Commands on Linux
Connect a multi-keyword model to a kiosk, robot, desktop tool, or local service with Python.
Custom Wake Word on ESP32-S3 with INT8 TFLite Micro
Run the tested ESP-IDF example, inspect measured latency, and replace the matched model and output head.
Custom Wake Word for ESP32
Train a product-specific phrase, choose ONNX or INT8 TFLite, and deploy it offline with ESP-IDF.
MultiNet Alternative for ESP32: 7-Language Multi-Keyword TFLite
Use one INT8 TFLite Micro model for multiple command words in Chinese, English, Japanese, French, German, Spanish and Hindi.
English Custom Wake Word
Type a word, get a small offline ONNX model in minutes. No recordings, no GPU. English, German, French, Japanese, Chinese.
Picovoice Alternative for Small Product Teams
Compare a commercial SDK, voice module, self-training and an online model platform for product delivery.
openWakeWord Alternative — Stable Recall, No Tuning
openWakeWord's custom models swing from 25% to 90% recall. Voicute's from-scratch TCN holds 90%+ — in English, German, French, Japanese, and Chinese.
microWakeWord Custom Models: Why False Positives Happen
Respect the strengths of preset ESP32 wake words, then fix custom-model failures with synthetic positives, real hard negatives and optional speaker recordings.
microWakeWord Alternative — Custom Wake Words Beyond ESP32
When a project needs multilingual models, cross-platform ONNX/TFLite export, or a managed custom-data workflow.
microWakeWord vs Voicute
A fair comparison of preset ESP32/HA wake words and a managed custom-data workflow for multilingual ONNX/TFLite models.
Stable 90%+ Wake Word Recall, Explained
Why frozen embeddings swing and from-scratch training stays stable — the architecture behind consistent recall.
The Best Wake Word Tools in 2026, Compared
openWakeWord vs Picovoice vs DIY vs Voicute — languages, multi-keyword, and the three questions to ask first.
Wake Word Training Platforms Compared
Online generators vs commercial SDKs, DIY training, and voice chips — output format, licensing, and the four checks before you pay.
Multi-Keyword Wake Word Detection: One Model, 10 Commands
Why one-word-per-model doesn't scale, and how a shared backbone detects 2–10 keywords in ~100 KB.
Multiple Wake Words, One Model
Every family member gets their own wake word, one firmware covers all SKUs, switching words is a config edit.
A Custom German Wake Word, No Recordings Needed
Why English tools fail on German, and how TTS synthesis + real-audio fine-tuning gets recall above 90%.
How Do I Train a Custom Wake Word for Home Assistant?
Train and test online, then deploy standard ONNX/TFLite models across Home Assistant, ESP32, Raspberry Pi, Android, Web, Windows, macOS, and Linux.
How Do I Train a Custom Wake Word for ESP32?
Train multilingual multi-keyword models and export TFLite for offline ESP32 voice control.
How Do I Train a Custom Wake Word for Raspberry Pi?
Build one ONNX or TFLite model for several keywords and run it locally with Raspberry Pi audio.
How Do I Train a Custom Wake Word for Android?
Train a multilingual custom wake word and integrate offline ONNX/TFLite keyword spotting.
How Do I Train a Custom Wake Word for Web Apps?
Run a multi-keyword ONNX/TFLite model locally in a browser with microphone permission.
How Do I Train a Custom Wake Word for Windows?
Export an offline multi-keyword model for Windows desktop applications and automation.
How Do I Train a Custom Wake Word for macOS?
Train multilingual phrases and run local ONNX/TFLite inference on macOS.
How Do I Train a Custom Wake Word for Linux?
Deploy one multi-keyword model to Linux applications, scripts, kiosks, and automation.