Bug: 156879015

Clone this repo:
  1. 1358c17 Mark ab/6881855 as merged by Xin Li · 8 weeks ago master
  2. 063ed97 Merge "Catch Encoding exception and ignore the exception line in logger." am: 8a0b78e594 by Jusik Chung · 3 months ago
  3. 8a0b78e Merge "Catch Encoding exception and ignore the exception line in logger." by Jusik Chung · 3 months ago
  4. da6a5cf Skip ab/6749736 in stage. by Xin Li · 3 months ago
  5. c6ab643 Defined a signal for the assistant launch under NGA scenario. am: 018a2264d6 by jschung · 3 months ago

OpenHST : Open sourced Hotword Stress Test tool

Hotword stress test tool is a python script to measure the Google Hotword performance on Android devices with simple test set up. The goal of OpenHST is making Google Hotword Stress Test tool as an open source project so that Android Partners can easily adapt on their development.

Overview

Hotsord Stress Test concept diagram

Prerequsite

Before setting up your HST test environment, please make sure to meet following preconditions.

  • DSP hotword capability on test device (android handset)
  • ADB connection between the test server and android handset
  • Internet connection on android handset
  • No screen lock setting on android handset
  • Install Homebrew (Mac)
  • Install pyenv for Linux and Mac.
  • Install GNU make(3.81 above) on test machine (Linux, Mac, Windows)

Test environment setup

Please follow below steps to set up HST on your test machine and make sure that no error popped up during the installation.

Linux & Mac

  1. pyenv install 3.6.8
  2. pyenv shell 3.6.8
  3. make start
  4. chmod +x ./start_venv.sh
  5. source env/bin/activate
  6. ./start_venv.sh
  7. make proto-compile

Windows

  1. install python3.7.5
  2. make start
  3. .\env\Scripts\activate
  4. pip3 install -r requirements.txt
  5. Download protoc.ext (3.11.4 version) from github
  6. Decompress protoc-3.11.4-win64.zip and copy bin/protoc.exe to env/ folder
  7. Decompress protoc-3.11.4-win64.zip and copy include/* under env/include folder
  8. make proto-compile

Voice model training

Before conducting stress test, voice model should be trained with the recorded tts sound.

  1. Please follow the guide in “Train voice model” chapter in OEM Hotword Stress Test with OpenHST

Stress Test

Please refer to the guide in “Run Stress Test” chapter in OEM Hotword Stress Test with OpenHST and run the hotword stress test.

  1. python stress_test.py --test_name dsp_trigger_and_screen_off --num_iterations 3000 --output_root test_out
  2. Over 95% pass ratio from 3000 test cycle would be fair enough to declare that the device passed the hotowrd stress test

Contacts

Join our mailing list for discussions and announcements.