About a year ago, the PatientsLikeMe Parkinson’s disease (PD) community started something totally different: a study to compare the sound of their voices to their self-reported PD Rating Scale (PDRS) on PatientsLikeMe. It’s called the Patient Voice Analysis (PVA), and we teamed up with you, Max Little, Ph.D. and Sage Bionetworks to get it done.
A little bit of background
Dr. Little had done some earlier work and compared the voice signals of people who were living with PD to those of people who were not, but we wanted to take that to the next level. With their PDRS, they shared how Parkinson’s was affecting them, and we were able to match their self-reported scores to the sound of their voice.
By matching a PDRS to voice samples, we might develop the ability to predict PDRS scores (which takes a few minutes to complete) by using the voice test (which only takes a few seconds). We might also be able to detect more subtle changes in people’s Parkinson’s through their voice than we can through the PDRS. This is what Dr. Little is working toward, and all the voice samples you donated will help make it happen.
Community results, starting with the basics
Who took part, from where, and how many PDRS scores could we match to voice recordings?
- Most of the recordings came from the U.S. (81%), with others coming from the U.K. (12%), Canada (5%) and some from Australia and New Zealand, too.
- 677 of you recorded 851 voice samples
- Since our original goal was 500 samples, you blew that out of the water!
- 114 of you took the test two or more times, and one community member even contributed 10 recordings!
- For those that took part more than once, we can start to examine how your symptoms changed over time.
Why voice recording quality matters
For the PVA study, you were able to use your landlines, cell phones, even Skype or Voice-Over-Internet-Protocol (VoIP) to submit your voice samples. The recording quality varies depending on which type of call you used, occasionally creating technical issues in analyzing the voice samples.
For example, if you’re using a cell phone in a busy restaurant, your microphone will automatically get louder so that the person you’re talking to can hear your voice better and without distortion. But that also changes the loudness of the background noise. In this study, that automatic change could actually affect the quality of the voice recording, so we have to identify where this has been an issue and find ways to overcome it.
Partly because of this, we’re still analyzing the voice samples in detail. We’re looking for subtle markers of Parkinson’s, such as fluctuations in volume, pitch and breathiness. We’re also training intelligent algorithms to identify when the quality of the voice recording is strong enough so we can develop a consistent and repeatable process.
You be the researcher
To give you some idea of what we are looking for in these voice recordings, we wanted to share a couple with you. The first is someone living with severe Parkinson’s, who scores 55 on the PDRS. You can probably hear the noticeable tremor in pitch, and the occasional short breaks in voicing.
The second is a recording of someone with mild Parkinson’s, who has a very low PDRS score of 1.
Can you hear the subtle drift in pitch? This is, most likely, indistinguishable from normal pitch drift. Subtle pitch variations such as this are one of the kinds of symptoms that algorithms attempt to identify from these voice recordings, and they contribute towards making the PDRS prediction.
So, what’s next?
At this stage, the PVA project is still just a research tool and isn’t quite ready for clinical or diagnostic use. We’re still working on analyzing the data to compare the severity of voice patterns to the reported severity of Parkinson’s disease. But in the meantime, if you’re looking to share more info with your doctor, the most useful tool is your PDRS score on your PatientsLikeMe profile. It contains items that make sense to a neurologist. If your clinic has access to speech and language pathologists, they would also have the ability to map any vocal issues you may be experiencing as part of your Parkinson’s.
As we continue to evolve the tools, we hope to provide individual level feedback and information for clinicians. But before that can happen, we want to make sure that the data quality is high enough to support drawing clinical conclusions.
Share this post on Twitter and help spread the word for Parkinson’s.