The most significant risk is the electroglottograph breaking down so that we can’t use it in our live demo, but luckily we already have plenty of recordings of us using it so our final demo will not be affected either way.

A calibration test we ran for our electroglottograph is using a larynx simulator to ensure that we could read CQ data properly so that way when we put on the electrodes we could guarantee the EGG was working properly. Since our project metrics are mostly based on feel and not quantitative metrics, we ran several surveys for different users on whether or not the EGG sensors were comfortable, whether or not our application was easy to use, whether or not our tutorial page was effective, and what parts were difficult to use.

The backend has have two main unit tests

1. Pitch Analysis Unit Test
   • Tests the pitch detection algorithm for 8 distinct recordings of octave scales
     • Includes keys: C4 Major, G4 Major, A2 Major
   • Tests the pitch detection algorithm for 4 distinct recordings of “Happy Birthday”
     • Includes keys: B4 Major, C4 Major, C4 Major (harmony), F2 Major
2. CQ Analysis
   1. Verify that the CQ graph output from a octave scale recording of both opera singers is reasonably within the ideal CQ range. This is largely an informal test since the intent is to check that the graph is just displayed as intended, rather than trying to verify that a CQ recording is “accurate”. We might still want to consider adding this in the final report as part of our testing section? If this is the case, maybe a “passing” test is that 50% of the measurements appear within the ideal CQ range since we expect trained opera singers to have CQs that are “near” the data for trained singers reported in David Howard’s research paper.

No changes were made to the design. No changes were made to the schedule.

This week I spent some time working through the tutorial pages of our program that will give the instructions to the user of how to use/setup our product as well as inform them about CQ and why it is important, as well as finishing up our project poster. I also conducted some testing/recordings and hope to get the feedback soon on whether or not the tutorial was comprehensive enough and if the CQ knowledge was informative. Overall, not a ton of work is accomplished due to other commitments but luckily they will end a few days before our final demo to give me time to debug and fine tune everything.

Progress is on pace, just need to touch up the tutorial pages and then publish it to the main program which should take less than a day, will start back up Tuesday after 1pm since that is when my last class ends besides capstone. Over the next week it is just gonna be helping Melina through the video as well as working on the final report + making sure we get all the feedback from the testing we conducted.

Schedule

Official Project Management Schedule Updates

• IN PROGRESS Backend Testing
  • DOING Pitch analysis testing
• DOING CQ change analysis testing
  • ‘testing’ in this context doesn’t really make sense, but I will be visually verifying that a trained opera singer has a CQ closer to the idealized CQ range than an amateur singer
• IN PROGRESS Inform about Change in CQ
• IN PROGRESS Final Demo Video

Pitch Analysis

• DOING: test all ‘Happy Birthday’ repertoires’ accuracy
• DONE: add function to output pitch based on the mouse’s position
  • If I am understanding correctly, the frontend can pass the mouse cursor’s “x_pos”, which would represent a frequency value, to the backend. The function I wrote, get_pitch_label(), takes this value and converts it to a pitch note for the frontend to display as a small label next to the mouse cursor
• Improved pitch detection accuracy to 99.49% for all scales

CQ Analysis

• DONE: add function to output values of ‘ideal’ female and male ranges
  • Instead of a function, I added constants (one for female, one for male) to proc_data.py
  • 
  • This data was “recreated” based on the research papers by Howard. The papers summarize this data, but don’t explicitly give values for the summarized data. I had to create this summary with exact values and with slight modifications to fit our applications needs. For female data, I calculated two means (one based on all trained female singers, and a second based on only trained female singers with 10+ years of training) to compute CQ ranges. For male data, I made an executive decision to use the last session of the M1 (Male 1) test subject. The last sessions represents the recording session in which M1 had the most exposure to vocal training.
    • Research papers consulted:
      • DM HOWARD D ROSSITER RESULTS FROM A PILOT LONGITUDINAL STUDY OF ELECTROLARYNGOGRAPHICALLY DERIVED
      • Variation of electrolaryngographically derived closed quotient for trained and untrained adult female singers
    • Recreated data (by me)
  • While this tasks is computationally done, I will be meeting again with Susanna to ensure that the way I formatted the data is appropriate for the frontend to graph. I’m only concerned that because our research is limited to CQ measurements of certain pitches, that we will have to consider this when making the graph area look understandable to the reader. We can either fill in the “gaps” with a linear model, or represent only the concrete data in a “candlestick chart” style.
• TODO: Confirm ideal CQ data format with Susanna for frontend

Final Demo Video

• I recorded the footage I need for the final video
• TODO: Record video script narration
• TODO: Edit and publish video