Posted on

Melina’s Status Report for 04/26/2025

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

Pitch Analysis

  • DOING: test all ‘Happy Birthday’ repertoires’ accuracy
  • DOING: add function to output pitch based on the mouse’s position
  • Improved pitch detection accuracy to 99.49% for all scales

CQ Analysis

  • DOING: add function to output values of ‘ideal’ female and male ranges
Posted on

Melina’s Status Report for 04/19/2025

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

Pitch Analysis

  • Adjusted intervals mapping frequency to pitch
  • Added backend functions for extracting frequency given a pitch (to help with creating graph ticks for the frontend
  • Tested pitch detection algorithm for all recordings of scales
    • Results were 97.9% accurate
    • DOING: test all ‘Happy Birthday’ repertoires’ accuracy
  • DOING: add function to output pitch based on the mouse’s position over the audio player view of the pitch vs time container

CQ Analysis

Posted on

Team Status Report for 4/19/25

End-to-end Product

We’re at a point where we are able to use our components to make a full end-to-end example of our workflow: starting with the recording of both an EGG signal and the corresponding audio signal. VoceVista processes both the audio file to get spreadsheets of CQ and pitch frequencies over the given interval. At the moment, we have to do much of the following processing manually in order to sync up the start times of each file, as well as the audio itself. Once these are processed and lined up, we can load them into our application. The backend will automatically make sure that the total duration is synced up, and establish the range of pitches to show on the display graph based on the pitch identification algorithm. Finally, the frontend displays all this data to the user. Unfortunately, this isn’t a particularly seamless process at the moment, since it still requires some manual processing of data, but it at least is effective.

New singer feedback

Our main collaborators in this project are the two opera singers in the School of Music who we’re working with. This week, we also completed brief sessions with three non-music-student vocalists, with the goal of further gauging how new users interacted with our application and setup. This was an extremely small sample size, but yielded some interesting results from the anonymous Google form we asked the singers to complete after their sessions. Notably, all three singers we worked with agreed that the sensors were comfortable to wear, and that the setup and the navigation of the app were straightforward. However, we also asked on a scale of 1 (strongly disagree) to 5 (strongly agree) whether “the information I gain from this application would be useful or informative to my practice as a singer.” All three of our singers reacted with 3 (neutral) to this question– they didn’t see a direct practical application of the EGG data, and seemed to expect a clear “this is good” “this is bad” feedback as to their CQ, when the reality is a bit more complicated.

This was informative to us because on the one hand, these singers were not as experienced in formal technique as our opera singers, despite their love of music, and this confirmed that our target audience is actually quite small: this application might not be the most directly applicable to a larger pedagogical setting. However, through various channels of feedback and the advice of our TA and faculty mentors, we’ve agreed that we need to do more with the CQ data that we’re gathering, since that’s really what makes our project unique. As a result, we’ve made a plan for displaying CQ data in relation to pitch, allowing the user to select multiple days to compare over time. See Susanna’s status report for more details. This will be the final main feature to implement in the final couple weeks before demo.

General Schedule

Our overall integration of components has been going according to schedule, and our EGG has not had further complications this week– there aren’t any new risks to the project as of this time.

Posted on

Susanna’s Status Report 4/19/25

Revisiting last week’s goals:

  • Use real EGG output data on playback page
    • Done– we took more recordings with our vocalists, and put the real CQ, pitch, and audio signal into the playback page– after taking a number of steps to synchronize all three of these files
  • Implement basic repertoire library page
    • Done– the library allows the user to mark each recording instance with the song it’s related to, the name of the recording, and the date. The final column in this table allows the user to open up the playback view for the given song. I implemented a sqlite database with two tables (one for songs and one for recordings, with a one-to-many relationship between the songs and the corresponding recordings)
    • Additionally, I created this “Add Recording” page where the user can create a new recording instance to add to this table, and can select the corresponding files to display. Additionally, the user can create a new song to link the given recording to.
  • Complete one data analytics view
    • Done – We added the view for seeing pitch on playback (as you can see under the first point), including the visual graph and the names of the pitches.

Next Steps

  • Pitch vs CQ view
    • The final feature to complete for the app side is a CQ over pitch analysis. Basically, the x-axis displays pitches, while the y-axis displays CQ. Multiple recordings from separate days can be loaded, to view the average CQ for the given pitch for each recording selected.
  • Testing & verification
  • Full integration with the backend

As you’ve designed, implemented and debugged your project, what new tools or new knowledge did you find it necessary to learn to be able to accomplish these tasks? What learning strategies did you use to acquire this new knowledge?

The main technology I had to learn for this project was, of course, the DearPyGui framework– I hadn’t programmed a Python desktop application since 15112, so I didn’t have much background to work off for this particular task. DearPyGui is a more obscure framework, chosen for its particular GUI features for things like graphing. However, this obscurity was a large downside to the framework, because learning resources were limited. I mostly used the DearPyGui documentation website, as this featured comprehensive explanation of components and functions, as well some limited examples. For more obscure glitches, I turned to the DearPyGui Reddit and Discord pages. Because of the small amount of resources, I had to do a lot of extrapolating from limited information to use in larger contexts. For instance, I wasn’t at first sure of how to create a multi-page display system within DearPyGui: I got information from these resources as to how to code the individual components for each page, but I created my own system for switching from page to page (creating a class for each page within the app, plus an app manager class) as the examples didn’t have that much detail.
Posted on

Tyler’s Status Report 4/19/25

This week I primarily focused on collecting and transferring all the data we collected in VoceVista into a useable excel spreadsheet for Melina and Susanna to use. Since our electroglottograph dual channel feature for the microphone and EGG signal is not working, I have to record the audio from a different computer, send it my computer, and manually align it with the EGG signal data.   On top of that, I have wrote up a tutorial for new users to follow to use our product, and am now working to implement that into our frontend. That has involved me learning a little bit of PyGUI but it has been relatively straightforward.

So far progress is on track, I want to finish up the tutorial page by this weekend and for next week be able to work through adding an overarching guide on how to read the CQ data as well as what takeaways there are from the data. So far, all we do is output the CQ data and leave the interpretation for the user, and while it is impossible to give hard truths for CQ data I want to write out a guide for general trends to see how you compare to the “industry standard” for opera singers.

Several things that I had to learn in order to complete our project is how to use VoceVista, how to use an electroglottograph, a little bit of excel spreadsheet work, and learn a few different python libraries such as pytesseract or DearPyGUI. The primary learning strategy I used for VoceVista was by using their online video tutorials to get everything initially set up and then emailing the author’s of VoceVista any miscellaneous questions I had afterwards. For the electroglottograph, I primarily relied on the manual that came with it but that led to several issues, but luckily I was able to get in contact with the manufacturers of the electroglottograph that we are using to fully utilize the electroglottograph. With the python libraries, since DearPyGUI is not well documented I primarily asked Susanna questions I had with using it since she had been using it all semester. With pytesseract I primarily looked at other coding examples similar to what I needed and followed their template of how to utilize the library.

Posted on

Melina’s Status Report for 04/12/2025

Schedule

Official Project Management Schedule Updates

  • IN PROGRESS Backend Testing
    • DOING Pitch analysis testing
    • NOT STARTED CQ change analysis testing
  • IN PROGRESS Inform about Change in CQ

Pitch Analysis

There have been delays in receiving most of the pitch data from this week for testing. One important recording, C Major Scale by one of our opera singers, was obtained and testing yielded ~97.83% accuracy. This task is a little behind, so I will be following up with the team to get access to more crucial tests, such as Happy Birthday.

CQ Analysis

Currently the program now extracts CQ values at timestamps that correspond to the time a given pitch is from. The next step will be to implement a way to choose a set of CQ data based on an identifier (such as the name of a repertoire or the key of a scale for “name-less” warm up recordings). The idea is have a system that allows the front end to pass in these identifiers and receive the clean sets of data to pass on to DPG for graphing.

I conducted interviews with the vocalists this week to gain a better understanding of what changes in CQ they would find helpful to know about. The main takeaway was that they would like help in identifying “flutters” and “wobbles”. I am currently working on understanding how a CQ measurement might indicate one of these problems. There do not appear to be research papers on this specific topic, but if we take CQ measurement recordings of these problematic samples and compare them with ideal samples, there may be a pattern we can discern.

Regarding the issue with collecting audio and EGG signals at the same time, I have proposed transferring ownership of the Voce Vista access key to my computer which has a USB-B port that may solve the issue. Unfortunately, this might mean limiting Tyler’s ability to collect data. For the time being, we have decided to focus on implementing the first upcoming task mentioned in this section so that the front-end has something tangible to demo for the vocalists.

Posted on

Susanna’s Status Report for 4/12/25

Revisiting last week’s goals

  • Use sample data to more clearly demonstrate the layered table rows in the playback view (for interim demo)
    • Done– there were some glitches here that I wasn’t expecting, but I basically just added a sine wave overlaid with the original audio signal. Tyler is working on syncing up our audio signal and EGG signal from our singer session on Thursday of this week, and once that’s complete I’ll simply use the average CQ data as the graph here instead.
  • Enable cursor scroll
  • Expand page menus and navigation
    • Partially done – This is sort of a vague task and I should have fleshed out more of what I was expecting it to encompass. I have created menus to link to the different instructional/playback pages, but since those pages aren’t completed yet, it’s not much of an accomplishment to note here.
  • Hammer down specific parameters for data analytics views, and complete one view (eg, average CQ over time graph) using arbitrary sample data
    • Partially done – Melina and I worked on establishing explicitly what we want some of our data views to look like, particularly with allowing the user to view multiple CQ waveforms simultaneously and specifically how to display CQ per pitch over time.
    • Partly still in progress – I haven’t actually coded the sample view itself. It’s a bit of a chicken and egg scenario between gathering data and coding data views, but I was hesitant to dive in head first as I don’t actually know what form the pitch data will look like once Melina has processed it.

Repertoire Library Wireframe

Additionally, I worked on the actual flow of the application as the user navigates it, with some help from our School of Music friends, since we realized that some of our initial plans were a bit hand-wavey. Specifically, how do vocalists view their library of repertoire that they’ve recorded and navigate it easily, especially given that this is a key element of what sets our application apart? My solution was to sort the repertoire at the high level by the title of the piece, and then at a lower level by the date each instance was recorded. Perhaps this is another case where it’s easier to see what I mean than for me to try to describe it. This is the Repertoire Library view, where clicking on each recorded instance will open up the Playback View that we know and love:

Verification

Our verification testing for the frontend will be done manually. My main goal is to ensure that the playback page functions given various parameters (files that have varying lengths, files that are missing either the EGG signal or the audio signal) and that unexpected inputs are handled gracefully with the app’s error handling. Specifically:

  • Input 5 songs of various lengths. Successfully play through, replay, and scroll through each song.
  • Input 2 songs without EGG signal. Successfully play just the audio.
  • Input 2 songs without audio signal. Successfully view the signal without playing audio.

Goals for next week

  • Use real EGG output data on playback page
  • Implement basic repertoire library page
  • Complete one data analytics view
Posted on

Tyler Status Report 4/12/2025

On Monday during class I was able to get a scraper in order for us to get the EGG data output available to us, I used the pytesseract library in order to do so. Tuesday I had a meeting with the manufacturers of the EGG and was able to debug and get the EGG signal and the microphone signal to work for the EGG. However, an issue that we ran into is that the EGG LED sensor for the electrode placement is broken, meaning that we will not be able to determine if we are getting the best electrode signal from the larynx placement while calibrating to the user. Another issue is that we are unable to utilize our EGG dual channel feature, which we believe may be because we need to utilize a USB to USB-C adapter in order to plug into the EGG. It is not a priority to be fixed, what we are currently doing is just recording using the mic on one laptop and recording the EGG signal on a different laptop and then overlaying them based on the audio timestamp of where we start. We start by pressing the record button together so it won’t be the most accurate but it will be useable for the time being. Now, I am working on merging the data sheets together in a nicely formatted way as well as recording more data from our singers, we had one two hour session with a singer but we would like to get a variety of singers before our meeting on Wednesday.

As far as progress, I believe that I am basically done with my parts and now I am focusing on helping Susanna and Melina with their parts and making sure our frontend design is intact.

For verification on my work, I have been comparing the measured EGG data with the EGG data that was provided as well as research papers to ensure that our EGG data collection makes sense and doesn’t output insane closed quotients that do not make sense. With my data collection of the EGG data, I roughly know it to be accurate since I am watching the closed quotient move through VoceVista and it matches the output of the data scrapper I designed. We do not need it to be super accurate since in the future VoceVista will be able to do it automatically, but since we do not have access to that version yet I designed a crude solution to use to meet our deadlines.

Posted on

Team Status Report 4/12/2025

The most significant risks for the data collection portion is the fact that we cannot get our dual channel EGG signal to work, so we need to use two separate laptops. The risk with that is we might misalign our data which will skew our data to be outputting incorrect CQ data during the singing. We are mitigating this by making sure we click at identical time stamps so that way we can easily combine the EGG signal and the audio signal with minimal offset. Worst case scenario if it is really bad we can design a clicker that will take the global time and click record at the exact same time so that we can eliminate the misalignment chance completely.

We have already been working with singers throughout the entire process, designing it based off of their needs and overlaying it in a way that they can understand the information being presented. However, we have been designing this with two singers throughout the entire process, so we will need to run through the entire project setup and usage with a new singer to see where they may need guidance and write up an extensive guide on how to use our tool or make certain parts more intuitive to use. We want to ensure that our project is simple to understand and easy to use for singers, so testing on a wide range of singers will ensure that we can hash out any potential issues the singers may have in understanding the data or collecting the data. Our main form of validation in these cases will be surveying both our new singers and our existing participants to ensure that we meet our original use case requirements.

We will compare our measured CQ data results with several different research paper ideals that we read on for differing genders and vocal ranges and see if they match the data that we got for verification that our data collection was  proper.

No changes were made to the schedule or existing design.

Posted on

3/30/2025 Tyler’s Status Report

This week, I was sadly unable to get much done because of sickness, but there is a lot to update on. For our electroglottograph data exportation, we will be able to export it similarly to our pitch data since the author of Voce Vista has told us that he can easily implement it through his end. He told us we can expect it by the next update, we will confirm an exact date soon (early next week). Another update on the electroglottograph repair, I am in talks with the manufacturer and have sent them a video of the problem on Friday, they are very quick about responding so hopefully on Monday I will have either a way to fix it or send the electroglottograph to them in order for them to repair it. I was also able to get on a zoom call with Professor Brancaccio to work on the electroglottograph with me but since she has a different model it was not too useful in terms of debugging our electroglottograph.

I think right now, we are all set in terms of Voce Vista and data collection, and all we need to do is get our EGG fixing and then we are set for final presentation. Outside of the EGG, I will help Melina with working through how to detect CQ change as well as how to display it to the user.

So far, roughly on pace, illnesses have thrown me off a little bit and made it so that we as a team have to make some slight adjustments to our schedule but still on pace. As for my personal schedule adjustments, after deliberating with my team, we decided against implementing a database since all of our data will be stored in excel spreadsheets in a unique folders for each user so it will already be organized for them to utilize.