What is the issue?

the app i am developing requires

• a coustomized hotword(keyword) detection
• korean text to korean speech
• korea speech to korean text

Hotword detection (Continuous Speech Recognition)

• Snowboy (KITT.AI)
  • customizable hotword detection engine for you to create your own hotword like “OK Google” or “Alexa”
  • DNN (deep neural networks)
  • can customize model(personal, private) from https://snowboy.kitt.ai/ (login, korean keyword is okay)
  • always listening (record thread can run on foreground service)
  • https://stackoverflow.com/questions/50956228/thread-within-service-android-app
  • no internet required
  • Light-weight (if there noise around, CPU usage may increase)
  • Supports various platforms and programming languages
  • Need a commercial license (Please contact snowboy@kitt.ai)
  • Android Demo https://github.com/Kitt-AI/snowboy/tree/master/examples/Android
• Porcupine (Picovoice)
  • Alireza Kenarsari (Founder at Picovoice) says porcupine has lower miss rate than snowboy
  • https://github.com/Picovoice/wakeword-benchmark
  • https://medium.com/@alirezakenarsarianhari/yet-another-wake-word-detection-engine-a2486d36d8d4
  • but when i really try it, snowboy worked better on my development environment (I may need to try more models)
  • AAR style library
  • Provides (Standard, Tiny) version
• PocketSphinx
  • demo: https://github.com/cmusphinx/pocketsphinx-android-demo

Speech to Text & Text to Speech (Korean)

• kaldi is a toolkit for speech recognition written in C++

• Zeroth Project (Kaldi based)
  • MoreCoin: a mobile app to collect voice data from various users link
  • Explanation of how korean speech recognition works: link
    • Multi layer structure to analyze voices
    • explain the difference between English and Korean
  • using AWS server, web crawler to collect 13GB pronounciation dictionary and language models
• KaKao API (Newton API) 20,000 requests free
  • provides both STT & TTS
  • reaserch of open source API(Korean): link
  • good quality of speech
• Naver Clova Speech Recognition API (STT): link, 4 Won/15sec

• Naver Clova Speech Synthesis API (TTS): link, 4 Won/1000chars

• CMUsphinx
  • How to use other languages? link
  • Need Korean acoustic model, language model
• Amazon Lex (STT) & Polly (TTS)
  • Maybe we can use third party language translation soulution
  • IoT solution: link
• IBM watson (SK NUGU use this API: http://www.newspim.com/news/view/20170206000145)
  • STT: link
  • TTS: link