Recording Task: Voice commands

Roger Jang (±i´¼¬P)


In this task, you need to record 24 voice commands twice to have 48 wav files, which takes about 10 minutes. The recorded clips will be used for further exercises in the class, such as end-point detection and voice command recognition. To proceed with the recording, please follow the stey-by-step instructions closely.

  1. If this is your first-time recording task, please read important notes about recording carefully. For this task, you need to use an omni-directional microphone, such as the one used in a notebook.
  2. Downlaod the recording program voiceCommandRecordingProgram.rar and uncompress it into a folder "voiceCommandRecordingProgram". Follow the "readme.txt" file in the folder: Please follow the steps to record 24 voice commands twice. Roger Jang, 20140316 1. Please downlaod the following two toolboxes: Utility Toolbox: http://mirlab.org/jang/matlab/toolbox/utility.rar SAP Toolbox: http://mirlab.org/jang/matlab/toolbox/sap.rar 2. Uncompress these two toolboxes and put them into two separate directories. 3. Change the "addpath" lines in "go.m" to add the toolboxes to the search path. 4. Type "go" under MATLAB to start the recording of 24 voice commands twice. The program "goRecord.m" will generate a folder "waveFile/dddddd" (where "dddddd" is your student ID, such as "921510") to hold all the recordings. Note that the "wave" folder should reside at the same level as the folder "voiceCommandRecogrdingProgram".
  3. After recording, you still need to manually label the end-points of the wave files. We need the manually labeled end-points as the ground truth for evaluate the performance of our computer program for end-point detection. First of all, you need to download the labeling program endpoingLabelingProgram.rar. After uncompress it into a folder "endPointLabelingProgram", follow the "readme.txt" file to perform labeling: Program for labeling endpoints Roger Jang, 20050314, 20060304, 20140316 1. Install the following two toolboxes if you haven't done so yet: Utility Toolbox: http://mirlab.org/jang/matlab/toolbox/utility.rar SAP Toolbox: http://mirlab.org/jang/matlab/toolbox/sap.rar 2. Modify the main program "goLabel.m" such that the variable "auDir" holds the path to the audio files to be labeled. 3. Type "goLabel" under MATLAB to start labeling endpoints. 4. The program will load each audio file. You can drag the red lines in the first plot to modify the endpoints. (Originally the red lines coincide with the magenta and green lines, which are the endpoints identified by the computer.) 5. You can hit ENTER or any key to save the file and move to the next file. 6. The endpoints will be recorded in the file name. For instance, 3a.wav will be renamed as 3a_849_31226.wav if the endpoints are 849 and 31226. 7. If you run into any problem, please get in touch with Roger Jang at "jang@mirlab.org".
  4. During the above process, do not manually change any file names or file contents, otherwise it will cause errors in the subsequent operations on file reading or performance evaluation.
  5. Please upload the folder "dddddd" to the ftp site specified by TA in the class. Any questions, please get in touch with TA directly.