DSpotter GarbGen Tool
繁體中文 ● 简体中文 ● English
Download
Release notes
Introduction
  • DSpotter GarbGen Tool is used to automatically generate GarbWord which may cause FA, and use them to fine-tune DSMT model.
  • To use this tool, please prepare
  • 1.Your DSMT file
  • Your DSMT file can have multiple groups, but the tool could only modify one group each time.
  • 2.Keyword script file
  • It's recommended that the wav files in script file should cover different genders, accents, and age groups.
  • At least 100 sentences for each keyword.
  • Please make sure the keyword in wav file is clear to hear, and the recoding background is relatively quiet so as to avoid the difficulty in recognizing keyword(s).
  • The script file must be in UTF-16 little endian encoding.
  • Each line of the keyword script file should contain absolute path of the wav file and the expected recognition result which can be a single command or multiple commands.
  • The absolute path and expected result should be separated by a Tab(\t). If there are several keywords, each keyword should be separated by a slash(/).
  • If there are several keywords, the order written in the script file(e.g. Hello/turn on/turn off) should be as same as the order in the wav file.
  • 3.FA testing script file(wav files w/o keyword)
  • The FA testing script is used to test the FA rate, we recommend that it should be at least 24 hours long.
  • The FA testing script file must be in UTF-16 little endian encoding.
  • Each line of the FA testing script file should contain the absolute path of the wav file, expected recognition result is not needed.
  • Please note that the size of a single wav file must not exceed 2GB.
  • All of the wav file should be saved in format: Sampling Rate 16000Hz, Resolution: 16bit, mono channel, PCM。
Login:
  • Please use DSMT account to login.
  • After login, you will see the UI as shown below.
Load DSMT：
  • [File] -> [load DSMT File] to load your DSMT file.
  • Find and open your DSMT file.
  • Check the DSMT file name and all of group(s) at the left after DSMT file is loaded, please choose the group you want to modify at the top right.
GarbWord Candidate
  • The first part of this tool is to generate GarbWord candidates using wav files contain keyword.
  • Load keyword script file(UTF-16): Load keyword script file(wav files contain keyword).
  • Load garbword candidate(UTF-8): You can also load the file GeneratedGarbWord_XXX.txt under default output folder if you have already run the tool ONCE.
  • Default output folder: (path_of_your_DSMT)\DSpotterGarbGen_Output\
  • Please note that content of GeneratedGarbWord_XXX.txt is based on the keyword script file. Different keyword spt file will generate different txt file.
  • Either "keyword script file" or "GeneratedGarbWord.txt" is required for this part.
  • If some of wav file(s) have illegal file path, there will be a warning message. Please correct the file path and re-load the new script file.
Output GarbWord
  • The second part of this tool is to pick useful GarbWord to reduce false alarm rate.
  • Max Num: Maximal number of GarbWord to pick. The number of GarbWord will affect the size of final model.
  • Load FA test script file(UTF-16): Load FA test script file(wav files w/o keyword).
Start
  • After the required files are loaded, you can press Start button and wait for the result.
Running
  • While running, you can click the [Stop] button to stop the program.
Finish
  • When finished, this tool will generate a new DSMT model(zipped), which contains garb words, under default output folder.
Export log
  • You can click [File] -> [Export log] to export log into text file.
Supported language
  • Arabic
  • Bahasa(Indonesia)
  • Bahasa(Melayu)
  • Cantonese(HK)
  • Chinese(TWN), Chinese(CHN), Chinese(CHN)-English, Chinese(CHN)_Tone, Chinese(CHN)_Tone-English
  • Czech
  • Danish
  • Dutch
  • English(AU), English(IN), English(PHI), English(SEA), English(SG), English(TWN), English(UK), English(US), English(Worldwide)
  • Finnish
  • French
  • Greek
  • Hindi
  • Hungarian
  • Italian
  • Japanese, Japanese-English
  • Korean, Korean-English
  • Norwegian
  • Polish
  • Poruguese(BRA)
  • Portuguese(EU)
  • Romanian
  • Russian
  • Slovak
  • Spanish(EU)
  • Spanish(LA)
  • Swedish
  • Tagalog
  • Taiwanese
  • Thai
  • Turkish
  • Ukrainian
  • Vietnamese
Download：
Release notes
Date Version Description
2024/03/28 1.0.10 Engine update, UI update
2023/12/18 1.0.9 Performance improvement, bug fix
2023/11/14 1.0.8 Update engine to 64-bit version
2023/10/24 1.0.7 Update DSpotter engine to 2.2.18.2, bug fix
2023/02/08 1.0.6 Fix a bug when file path of a wav contains wide charater
2022/11/25 1.0.5 Add CHN Tone Model, bug fix
2022/10/28 1.0.4 Bug fix
2022/08/10 1.0.3 Bug fix
2022/06/06 1.0.2 Support multi-window, support multi-group, bug fix
2022/01/04 1.0.0 First release