BeSpoke voices: Empowering personal and social self-identity for those who rely on speech generating devices to communicate

VocaliD’s breakthrough speech technology is the focus of this webinar presented by Dr. Rupal Patel - Speech Scientist and Technologist. BeSpoke Voices leverages years of research to create custom crafted voices by blending only 2-3 seconds of the recipient’s voice (the Vocal DNA) with recordings of a matched speaker from its crowd-sourced Human VoiceBank. BeSpoke voice captures the clarity of the matched speaker but the vocal identity of the recipient so s/he can be heard through his/her own voice. (Get the slides.)
 

 

 

Transcript: 

- [Anne Marie] Great good afternoon everyone. It's four o'clock so we're going to be getting started. Thank you for joining us for this CTD Webinar. We are pleased to welcome Rupal Patel Ph.D. Speech scientist and technologist. She is going to be discussing BeSpoke Voices, empowering personal and social self-identity for those who rely on speech generating devices to communicate. I'm going to go ahead and pass it off to Rupal so that she can go ahead and get started. Thank you.

- Good afternoon everyone. Thank you for joining in. My name is Rupal Patel and I'm the founder and CEO of VocaliD. And also a professor at NorthEastern University. I'm going to have a... I'm going to turn the webcam off in a few minutes as we go through the slides and then we have time at the end for questions that I'd love to take as well. So I'm just going to go ahead and turn this off now and switch over to the slides. Okay. So as I said I am the founder and CEO of VocaliD. And what I want to talk about today is the science and underlying technology that was initially developed in my laboratory in NorthEastern University where I'm a professor in the department of communication sciences and disorders and computer and information science. And that gives rise to VocaliD. And the underlying technologies of relevance to assisted technology which is what we're talking about today. I want to delve deeper in the science ofProduction and how that impactsmakes custom voice for those who rely on assisted communication devices. I'm very excited to announce that we will be launching our spoke voices as well as our vocal legacy voices later this year in 2016. And I'm going to tell you how we create these custom digital voices. So each one of us has a unique voice. It's how people know us and it's how people remember us. All of our voices are not identical and in fact they are so much a part of who we are. They're our identities. And there's an evolutionary basis for why we all have unique voices. It may have begun with the fact that we needed a unique voice to signal danger or to even make mating calls to identify who it is amongst our species that we are communicating with. Or those who are outside of our species or group. Right. So there's a very early key reason for why all of our voices are different from one another. And it goes far beyond the technology today that we're talking about. It really has a deep meaning in our emotions and development. But the very basic level, a unique voice allows us to identify the person that we are speaking with. And a familiar voice actually activates different parts of the brain. And a mother is able to and fathers are able to identify the voice of their child from a sea of cries and voices or chatter. And so there is very good reason as to why voices are differentiated. The human voice is seen as such an important intricate part of self-identity. And there are also a number of cultural and societal biases and norms that are associated with voices. We know from just a single hello over a telephone line that the person is that's calling. We can even predict their speech and gender and what they may do in terms of their occupation or habits and personality just from their voice. And that's such a power piece of how we make connections with ourselves but also with those we know and love. Yet with all the advances in artificial intelligence, voice activation that we hear about today, a compelling voice interface is really still sci-fi. You'll hear it in, or you'll see movies that just where the characters are artificial characters that have these unique voices that people can even fall in love with. Yet we all know, for those of us who are using assisted technology who have family and friends using the technology that the voices on those devices is nowhere close to the sci-fi future. And that's actually a real paying point for over ten million people who are living with voicelessness and using these speaking devices to interact with the world around them. And so now we've giving them these uniform voices that don't fit their body or their personalities. And I think if you think about the fact that it's just a technology has opened doors for people with severe communication impairment it's a significant leap forward. The fact that we have AAC is a significant leap forward from not having it and that's only in the last 100 years. But given the leaps in the medical model the focus of AAC up until now has really been on addressing the physical limitations of voicelessness. Now there's an increased focus on the social model of disabilities and it's important for us to start considering some of the personal factors. And a voice being one of them. Now in the medical model it was fixing or trying to fix the physical limitation of not being able to speak clearly. And so giving someone a functional voice solution is important and it sort of fixes that piece. But as you think about the social model, the social model considers other types of personal factors as well, the voice being a very important part of it. I think it's only now that both technology's advanced and cultural expectations of how we should be thinking about the severity has also advanced that we can think about the entire aspect of the individual not just their disability but how that can actually be seen as a feature rather than a How do we create ways in which these individuals can have full lives and be accepted fully for who they are and if we can harness some aspects of their voice perhaps that allows them to engage more fully in society and also see themselves as full members of society which they are. And so to only think about AAC as a successful transmission of information and basic information is to shortchange that person of their individual life situation. There are a few children who are using AAC in the classroom, they have similar typical cognitive abilities. But they shouldn't, and they may be even using the same device but they a unique individual to have different voices. And so one of the things that struck me is the beginning of VocaliD back in 2002 I was at an assistant technology conference presenting my research in how people with severe communication impairments still have control over the melodic aspects of voice. Socommunication partners can typically understand even most severely impaired speech. And that's because there's some aspects of the voice that are still relatively preserved in severe communication disorders. And that's the work that I had done for my PhD showing that there's preserved prosotic content prosotic meaning the melodic aspects of speech are so easily controlled to a relative degree by people with communication impairment. But yet right after giving that talk I walked into an exhibit hall and what I saw was a young girl and a grown man having a conversation using different devices but the exact same voice. And what struck me was that I was hearing this voice from all around me. That perfect tall voice at that time coming from everywhere. Hundreds of people using the same voice. Am I hearing an echo? The implication really of a sea of uniform voices... I don't know if I should turn off the audio here, you all hearing echoes? We're going to stop for a second. Okay, that's better? Alright. That will help us here. So in 2002, with the sea of uniform voices what I saw and heard around me was really something that sparked the beginning of VocaliD and the beginning of a career, a research line in this work. And I think that the main thing that struck me was that having these uniform voices really impeded social integration for people who use assisted communication. It impedes the ability for us to see the individual with the communication impairment as more capable rather than just looking at their disability. What I wanted, what I was seeing in our research is showing that even though the communication impairments had abilities that were preserved and that was the concept yet thedevices that we were giving them is the voice was still perpetually that disability. And I think it's not a mystery why sometimes adoption of AAC, even though it gives us so much in terms of being able to communicate, why adoption rates aren't as high as we want them and why abandonment rates are so high. And I think that part of is we have fullyon what this assisted technology should be doing for the individual. It should be giving them something beyond just that functional solution. And so what set off there was the beginning of VocaliD back then. So in 2002, what I saw was something that really sparked the beginning of the next line of work that we would want to do. Some other reasons why that similar voices were being used on assisted technology system was because of an artifact of old research. Back in the 1970's and 80's it was shown that the adult male voice is the easiest to understand in reverberate voice, reverberate environments like a classroom. But classroom acoustics changed over the decade and yet clinical practice fits people with the same adult male voice had not changed. And we needed to change that. We wouldn't dream of fitting a little girl with a prosthetic limb of a grown man. So why would we give them the same prosthetic voice? And really I think what are their options? Right now, the options with assisted technology needs really range in these two broad extremes. On the one end there's these low quality, robotic voices that are based on a model based aspect of speech synthesis. And we don't really have that in AAC anymore but we still hear that in the voices around us in other things that talk. And on the other end of the screen are expensive generic sounding voices like Siri or Cortana. Our voices in AAC fit somewhere in between. Yet they still are very generic. To have about 70 voices across multiple languages and only a few options for kids really isn't sufficient to mark all the various different people that are using AAC all around the world. So that's where we wanted to innovate on. Is we saw that there was an opportunity here to move beyond what we had then. Okay, so how do we even begin? How do we begin to build custom crafted vocal identities. Well, because we as a speech scientist, I'm trained as a speech scientist and a clinician, there was theory of speech production called the Source-Filter Theory of Speech Production. So we actually leveraged the Source-Filter theory of speech production to a great degree to create our custom voices. And the Source-Filter Theory of Speech Production states that speech is a combination of the source which are the vibrations of the vocal fold that makes sound. Once that sound is then pushed through the rest of the vocal track which we call the filter, these are the chambers of your head and neck that take that sound that almost sounds like an electric razor, you take that sound and you turn it into consonants and vowels and you shape the tube of the vocal track into different shapes. And that's what speech is. It's a combination of source and filter. But there are also a number, the people who have severe communication impairment have the ability to control the source of their sound but the filter is impaired because they can't move their lips and their tongue adequately to create clearly discernible consonants and vowels. So what we do at VocaliD is you separate the source and the filter and take only the source characteristics of the person who is unable to speak and borrow the filter characteristics from someone who is similar in age and size and cultural linguistic background. And when we combine them again what we get again is a voice that sounds as clear as the surrogate speaker but is similar in vocal identity as the person who we want to create the voice for. So that's really the science behind VocaliD. And in 2014, I was able to talk about this work on a relatively big platform at TED. And that led to a lot of people saying they wanted to contribute their voice to help people who were unable to speak. So I formed VocaliD so that we could take the technology out of the laboratory and into the real world so that we could create voices beyond just a few dozen that we could do in the laboratory. And I took leave from the university in 2015 to just jump in full time to create this company. I began by recruiting the best talent so that we could bring this vision to reality. I recruited first Geoff Meltzner a leading speech scientist, speech engineer with industry experience. Geoff and I have also collaborated on work in silent speech recognition. And also Michael Suen. Michael is a talented UI/UX designer and he is really the brains behind a lot of the interface that you see in VocaliD in terms of the voice recording platform as well. Our team is growing and we have a number of people who are entrepreneurs as well as engineers who are helping us create the technology as well as getting the word out to the people this could benefit. So what we first did in VocaliD is we created a platform for people to be able to contribute their voice. Because this was an important piece about scaling the technology. In the laboratory we had a few dozen voices that we'd recorded and from that we were able to create these unique voices. But if the voices aren't similar to the person you want to make the voice for you don't get a very good match in terms of how the voice is going to sound. So we wanted, we had this vision of getting the voices from people from around the world people of different accents and ages and groups and so on so that we could have a variety of voices from which we could create unique voices. So the voice recording platform looks much like what you're looking at now. It's virtual recording studio in which there is a console. And on that console there are a series of sentences that have been specifically designed so that we can capture all the sounds and sound combinations in the language. And as you say these very different sentences there's a projection on the back wall that shows the different consonants and vowels. And those consonants and vowels are those that make up the English language. And as you say each sentence, different sounds light up and he banks over time a variety of the different combinations of the language so that we have all the sound combinations so we can create sounds and words that were not said before. So this is the human voice bank platform. And from this platform we have been collecting a variety of voices from people all over the world. There are currently 12,000 voice donors from around the world who have been contributing. This map that you're looking at are where these people are from. These are everyday people who are wanting to contribute their voice to help those who are in need. And you can see a variety... Most of these people are coming from North America and Europe and parts of Australia but also lots of other parts of the world. We are currently only in English but we have been trying to get more and more people to contribute even with accents so that we have a variety of voices. We've got people on the platform from six all the way up to 91. So there's a huge range here and it's great because it allows people to create this whole amazing texture of the voices that we can create. So just kind of wrapping back to the technology part, what we need from someone who want to create a unique voice in away is you take two seconds or so of the sustained sound and from the person who you want to make the voice for then you search through the database and find somebody who's closest to them in age and gender and culture and location who has recorded about six to eight hours of speech through that voice banking platform that I just showed you and you can combine the two together to create this unique voice. So there's two real kinds of ways you can make a voice. For someone who is unable to make much speech at all or isn't able to speak because they have a severe communication impairment we call those voices BeSpoke voices because we're going to take a short sample voice and combine it with someone in the voice bank. And let me just play you a couple of samples of an individual where this is what we got from them in terms of their sound sample. Can you play that sound Anne Marie?

- [Anne Marie] We won't be able to play it through the PowerPoint, no. Do you have it on a file?

- [Rupal] I can open it on a file. Let me just do that. Okay, so here's the individual's sound. So that's all... that we were able to get from this one person who is an AAC user. She's a young girl, she's nine years old. And here's a voice sample of someone who was a match speaker for her that we found in the voice bank who we're going to combine that sample with hers from.

- [Girl] Even your voice can change like

- [Rupal] So now you've got these two voice samples. As you blend them together and you create a synthesized voice you get a voice like this.

- [Girl] I feel happy. My name is Mave.

- [Rupal] Okay so there's a variety of different... So what we've done here is taken the sound that the young girl can make and blended them with somebody who is similar to her in age and gender and created that synthesized voice. So that's what BeSpoke is. What a Vocal Legacy is is if you are able to bank your own voice then we can create a voice for you from your own recordings. So here, now I'm going to play you a sample of somebody who has been able to bank their voice prior to them losing their voice. So this is an individual who, you'll hear the original recordings and then you'll hear the voice that was created from those original recordings. So here's the original recording.

- [Man] What have you been doing?

- [Rupal] And notice his accent.

- [Man] Tell me about your family.

- [Rupal] Okay. So those are the original recordings. And now I'm going to play you the sample of his voice that we have synthesized through our Vocal Legacy process.

- [Man] Hello. This is my new voice. The quick brown fox jumps over the lazy dog.

- [Rupal] And so that's how we create the vocal legacy. If there's enough speech from us you have from the individual who one makes their voice for. Okay, so I'm just going to go back to this presentation again. Hopefully that gives you an idea of the way the voices are created. And at the end of the presentation I'll play you a video of the process as well. So we have currently seven people who we delivered our voices to. These were our trailblazer users. We created these seven voices in December of 2015 and here are some of the testimonials we have from those people who have been using their voices since then. For one individual, her therapist has noticed an increase in pre and post measures of the voice use. And what we saw is that prior to her using her personalized voice, the level of communication initiation and participation was a lot less that it is now. It is now 300% more. In other words if she was raising her hand two or three times during a session, she's raising her hand upwards of 18 to 19 times now. It's an incredible amount of change. And I think it was 6 to 18 was the difference there. Other things that when the first time for example Sarah heard her voice she said it sounds funky. These are not the kinds of words that people often use to talk about their assistive voice. Max's father said that when he heard his voice he heard his son's voice he said oh it sounds like Matt. It sounds passionate, just like he is. A gentleman named John who we made a voice for who had lost his voice to ALS who also went through the BeSpoke process with, when he first heard the two samples of the voice we created, the first couple he politely sort of nodded and said yeah that sounds like me. And the third sample that we created his entire body went into shakes indicating that he was hearing himself again for the first time in eight years. So I think it's really powerful what the voice can do. It's clearly still a synthetic voice, we have to keep that in mind that with all the advances we're making, we have been able to create a voice that's from just a couple of samples of their voice and this price point but it doesn't mean it's going to sound like an impersonation of someone. So just to keep those things in mind as well. We currently have 86 pre-orders for 2016. And are taking additional pre-orders for people who are wanting to build a voice for themselves or for their loved ones. So so far what has gotten us to this point is money through the... Or funding through the National Science Foundation and the National Institute of Health. In the allows us to take the technology which in the laboratory is taking us somewhere between 40 to 50 hours to create each voice, obviously prohibitive in terms of scaling them. Also hundreds of thousands of dollars, or at least tens of thousands of dollars to produce. In one year we've been able to bring that down to somewhere between six to eight hours of manual time. It's still expensive for us to build voices yet it still makes it far more achievable in terms of accessible to the public. It's a new technology though, and every new technology there is a time period where we need to have early adopters who want to support this and want to see this come to fruition. That last year we ran an IndieGoGo campaign for many people who contributed to that and also helped us get that first funding together to create the first seven voices. And we've been featured in a number of really exciting press venues that have helped us tell the story. Mostly for us to get more people donating on our voice banking platform but also to allow us to let the technology be known to more people around the world. We've been given some awards and have mentorship through different kinds of organizations including accelerator programs such as Mass Challenge and And now we're at the point where were are about to launch the product and also are seeking outside investors to launch the company. I think I'll take that as a wrapping up to this and I know that there are going to be more questions in the chat, and I've got a few questions here that I have listed in. So I'm going to take some of these in order and then what I'll do is go back to the chat and see if there's additional questions that don't suffice in terms of the answers that I have for these. So the first question I see on my list here is how much does a VocaliD voice cost? And so gonna put up this slide right now. Currently a VocaliD cost $1249 for us to build to voice and we have an annual subscription that's $240 to allow us to continue to be available on the various different iOS and/or Windows platform that are constantly evolving. It's costly for all the changes that are happening. In addition we have plans for ways in which we can evolve the voice over time as well. So if you built a voice for a nine year old today how do you grow that voice for later. So we partnered with some key assistant technology companies these companies in an initial starting point but are looking to have additional partners in this group as well. So the voices will work on a variety of platforms. I hope that answers some of the questions around this. If you have other questions around this please email us at hello at vocalid dot C O. The next question here I see it was what type of funding sources are available to fund the voice. And what we thought initially, currently our voices are an out of pocket expense. And we are working on ways in which we can get insurance to cover this but insurance is a bit of a chick and an egg in the sense that you need to make sure that you have enough proof points about the technology before you can get that kind of coverage. It's years, and by that time we have to actually prove that there is a market for this and that there are people willing and interested in this technology. So we're pursuing insurance but at the moment we need to have support from people who believe this technology needs to exist. And so it is an out of pocket expense. It is $1249, we priced it at a point where we're just trying to get people to use our technology. And what we've found that the online fundraisers that we've helped people do, either through GoFundMe or IndieGoGo and a variety of different methods people have been able to raise money very quickly. So in fact Max's campaign that raised over $10,000 was done in 22 hours. And so Leo's campaign was also very quick for the $1249 dollars that he was trying to raise. So there are many organizations, community organizations that can help. for example has actually helped us fund voices for people in the community as well. So there are a variety of different ways in which people can get their voices. Can get the funds for the voice as well. In terms of one other question I see here is on how can I help spread the word. So we have a number of people this summer that we are accepting into our fellowship program to help us get more people to contribute their voice. So when someone contributes their voice they can get certificates of appreciation or completion that can be used for community service hours. So there's a variety of different ways in which people can help us spread the word about voice donation as well as having people get voices who need voices. So I'm going to turn the web cam on at this moment and see if I can answer any other questions that people may have. Let's see. Ann Marie I don't see the web cam again. There we go. Okay, a couple other questions I have from here is my computer may not be the best device to get clear recordings. Are there recording studios that you're partnered with to get higher quality recording. We aren't partnered with other recording studios but there are many people who are recording in a home studio or sometimes even recording in a quiet room, or even a walk in closet can work relatively well. As long as your computer has a, you can put in a headset microphone that usually is very effective. Another question I'm seeing here from Jackie, can you clarify what it means by capturing an individual's voice quality. Sure. An individual's voice quality, there are many things that make our voices unique. So the pitch of your voice, the loudness of your voice, the raspiness of the voice, all of these are actually signals that are produced by the larynx so there are ways in which we capture when we record someone's sound just saying aah or saying aah production, we actually get a lot of information about their voice from that. I hope that answers your question, Jackie. Why do you think Stephen Hawking... Ken asks why do you think Stephen Hawking still uses a speech device with a voice that dates to the earlier days? I think for Dr. Hawking I think the thing is that voice has become his voice. It has become a part of his identity and it's hard. And so it's hard for him to change and doesn't want to change it. And there are going to be people like that. So I don't think that every person who uses a speech generating device is necessarily going to want a custom voice because they started associating that voice that they've been using with themselves. But should they have been given a choice earlier in life about having a custom voice that probably had been an option they would choose because they wouldn't sound much like everyone else. In fact John and Linda have a very interesting story about this. John has been using the Ryan voice for the last eight years on his assisted communication device and they talk about it as being a third person in their relationship. And that when they were waiting to get their VocaliD voice, it was really about regaining that connection to one another not having this third voice in the mix of all of it. So Joy asked a question that are you aware of any research studies that discuss adoption rates of AAD. Yes Joy there are some studies on the adoption rate of AAD but they're not, we don't have concrete reasons for why people don't want to adopt or abandon the AAD devices. What we don't know is what are the reasons and also if you don't know a new technological trend that are coming down the pipe, then it's hard to predict why someone hasn't adopted that particular, the current state of technology. Let me see, I'm just going to write down... So Jessica asked this question will users custom voice change or be blended with different donors as the user ages? That's a great question. What we think about the voice as having four stages. A child stage, an adolescent stage, an adult stage and a mature stage. And so within those bounds we believe that there's ways for us to transform the voice. And that's some of the R and D that we're investing in heavily right now is how do we make those voice adaptations. But as you know, as someone goes through childhood to becoming an adult or a teen, there's a huge jump. So in that huge jump, it may be the same person, or it may be a different individual. But what we want to do is have that be seamless. It may also be that initially they chose someone whose voice was soft and similar to theirs, but as their own voice changed, they preferred a voice that was deeper or louder or whatever. So sometimes I think preference plays just as much of a role as fit. So one of the things we're doing now is when we create the voices, we learned some really important lessons in delivering our first seven voices. What we learned is we created three voice options for each individual in our first set. And the reason we had done that is because we didn't have a way to know exactly what in this that they would want, so we created these three different options and we played them the voice options and sometimes they prefer the ones that were not necessarily the clearest but were the ones that somehow they affiliated with. So now what we're doing is once a person has given us their voice sample we find the match with a match in our database, and we offer them one of three different choices. We are going to offer them one of three different choices of who they want to be blended with. So it's not only their voice but it's also their choice about who they want to be blended with. That's very important. The other thing we've learned in the first set of trailblazer voices is that sometimes the voices that are chosen by individuals who use AAD tend to have a smaller speaking rate. So you and I may speak at fast speaking rates and may prefer faster voices, but those who use AAD devices may have auditory comprehension impairments or may just prefer the slower voices. So everyone has a difference in their preferences as well as a sense of what fits them. Let me see if I can find another question here. So there's another question here do you only use recording for one match speaker or several? Currently our process is to use recordings from just one match speaker. There are interesting ideas there in terms of how you seek data from multiple people, but what you hear is jumpy sounding voices and so our goal is to make a voice sound very clear and understandable but also authentic. And that's actually a relatively hard thing to do with very little data from the recipient and also in relatively speaking very little data from our voice donors. So Siri was made with hundreds of hours of speech and costs millions of dollars to produce. What we're trying to do is taking hours of speech from someone who is non-speaking and have them... Someone who is able to speak and have them create a voice that is understandable and that's unique to them. So we're trying to push the envelope on the amount of speech we need and how quickly we can make this happen. Tom I see a question from you about partnering with NPR and their affiliate stations to capture voices. So in fact I'm giving a talk the other day sorry doing an interview at WGU here in Boston and we're talking with the sound engineer to figure out other ways in which we can use booths and booth time in the down time. And in many places some of our volunteer organizations that we've been working with have a lot of access to the university booth or to the audiology booths in different private practices. So the more sound recording studios that we can use the better, but quite honestly technology for the recording has gotten so much cheaper and better that even a cell phone, although we don't have our voice recording samples, you don't gather them off the phone and the reason is that when people are on the phone they tend to do this kind of thing, the microphone's on one side and what happens is as you continue to locate your phone or change the distance of the phone to your mouth you're actually going to change the clarity of the recording. So we ideally, if you can record in a sound recording studio, that's great. But that's not going to be the breaking factor in terms of us creating a high quality voice today. We're seeing that even people when they record from home with a headset microphone in a quiet room can have a good quality recording but it does need to be quiet. And it does need to be done in a way that complies with all the things that we ask people to do. Okay. So Dr. Milligan you ask how can I identify people who need help in my area. Do you have a list of people who need voices and can't afford them? Currently we don't have a list of people who need voices and can't afford them and if there are ways, you can always have those people get on our wait list and indicate that they are funding for themselves, in other words would want a voice but can't afford one right now. During our IndieGoGo campaign we ran another campaign if people had financial need we were able to help them along in that way. What we're doing right now at scale we must show that we have enough customers to get this off the ground. So there are many other funding vehicles available so we hope that if you do know of people please let them know about it and see if there can be any community based organizations that we can get to help us get this moving along. Ashley. Ashley asks to donate a voice, do I need to purchase a special kind of microphone. Yes Ashley, you don't have to purchase a special type of microphone but there is a microphone that we recommend. And that microphone is a Logitech microphone. You can find this information on our website. And it's also if you need more information you can find that at the hello at vocalid dot C O and we'll send that to you too. Okay let me see I'm trying to raise some of these other emails, or chats. How do people... I think I already covered this question about how do people access more personalized voices I E Medicare, Medicaid. Okay Shannon asks for Legacy he sounded English, pre recorded then more Australian after. So Shannon recognized a difference accent before and after of the gentleman. He's actually Scottish so he'd be somewhere in between. It's not about vowels distortions, but we're trying to re-create his speech from just 3500 sentences. So there are some approximations of speech that are happening there. Which is why I was talking earlier about the fact that it is still synthesized speech. But it is one where we're getting closer to the likeness of the individual. Okay. Soasked if a recipient has old recordings of their voice and can no longer speak can old recordings be used? So this question actually came up yesterday as well. We have a number of people with degenerative conditions who did not bank their voice and who are unable to actually even make sound now in the case of many people with ALS. And our current process requires that either we have a two second long sample which we're probably not going to be able to get at this at asample like that in a pre recorded sample of like a telephone record like a message recording or a video or something like that. Or requires us to bank to voice on our platform. There's ways to do it. And in fact we have applied for some funding in order to try take that what we call found audio and use our same processes on it. But we don't have the funding currently or capacity to actually do that although we would love to do that. So if we were able to get additional funding, that would be one of the first things that we would attack that how do we take found audio and use theprocess on that. I think it would be wonderful for so many people. It really right now we have funding through the government for the R and D necessary to really get these first two products out which is a lot on our plate with an extremely dedicated team. And we can't tackle all the problems at once and we need significant funding to take on these newer kinds of R and D. But none of these things are really rocket science. They're within our reach. And we're capable of doing it. Let's see. We still have another 15 minutes or so so I'm just going to check down because some of these questions... Have any member of the deaf community expressed an interest in receiving a custom voice? That's asked by few. There are actually some people who lost their hearing later in life but this is text to speech, and so I haven't heard people who have wanted a voice that have lost their hearing, but there are a number of people who have lost their vision later in life who have expressed an interest in wanting a custom voice. Because they rely ontechnology and currently evenleaders have very generic sounding voices and so what some of these people claim is that when they hear their voice, when they hear their own writing from a screen reader that is a generic sounding voice, they feel like they don't have that inner writing voice. And so they prefer to have a custom voice instead. No our e-mail address, John asked did you leave off an M in your email address or in the company address. No, our URL is of a company so it's VocaliD dot C O. Dot com was taken so we are dot C O. Many start ups actually have shorter or different kinds of URL's, so if you're emailing us, or if you're looking for us in the website it's VocaliD dot C O. Candace says there are many students at school that could really benefit from this technology for communicating their ideas. That's wonderful, Candace and if you can help us get the word to those students and their families it'd be wonderful. One of the things that we've been doing is working with some special education schools and organizations that work with them to see if we can do some, a number of pilot studies or pilot programs to get a number of kids in that school to have a voice. So here in the Boston area there's a school called the Cotting School that we've been working with that have our custom digital voices. And what we see is really the interaction with those kids with one another. Those students with one another is just fascinating. Two children in that class or two students in that school one nine and one 16 had the exact same voice and now they all have different voices and that's really has been very empowering for them. So I think I've gone through all the other questions are there any other further questions?

- [Anne Marie] I think we have some more questions, can voice donors be family members such as siblings that the individual has limited speech sounds?

- Oh sorry I didn't scroll down. Tell me that question again.

- [Anne Marie] Can voice donors be family members such as a sibling as the individual has limited speech sounds?

- Yeah, so a voice donor can be anyone. And although sometimes siblings, people are motivated by helping their sibling I think that's wonderful, I think siblings aren't always the best match because vocally the lung capacity of someone who is unable to speak or even the posture sometimes because if they're in a wheelchair, those things will affect their voice. And so although siblings are great, they may be a match with somebody else but even identical twins don't often have the exact same voice. I think we have this preconceived notion that thinking that siblings, people in the family all aren't going to sound the same because we got confused with our brothers and sisters growing up, but it's not always the case that the sibling will necessarily be a good match. And that's why our algorithms we'll obviously but that in the sibling recording if we have them will be in the pool of people from which we can select. But we look at a bunch of different characteristics of the match. Some things include gender, age, location and so on but also voice quality. So we have created a voice taxonomy that is a variety... Everyone has a unique voice but we can kind of group them into a set of different voices types based on pitch and loudness and breathiness and nasality or the locus of resonance. And these helps us figure out what some of the best matches are for an individual. So it's great to get siblings involved if they think just beyond just getting them to do something for their sibling who has a assisted communication needs, it's a great way to build empathy but they may not necessarily be the match.

- [Anne Marie] I see another one here. How do you address prosodic features of speech is that possible or will the voice still sound relatively monotone?

- Yes that's a great question. So address prosodic aspects of speech, these are the changes in pitch and loudness and intonation, that's really the basis of my research and a lot of what we are finding is that the prosodic aspects of someone who is non speaking are actually relatively reserved. So we capture those but you'll hear that synthesized speech still lacks intonation. It's what we're working on, we have some visual spacial waves in which we can indicate prosidy that we'd love to build into the voices but there is still work to be done for the emotional aspects of the voice. So Step one let's get it to making it sound authentic, step two and clearer still step two let's work on the intonation and the emotional aspects because that's a little bit more slippery. So Deborah I see a question from you. While recording my sentences I sometimes get warnings about volume ecetera and messages say not to worry, we will re record the problem sentences at the end so how much of your recording is needed? So we have some online methods for how we are testing and making sure that your recordings are of the quality that we need to create a voice. You need to have, because we're crowd-sourcing this, we have people recording anywhere and sometimes people forget that they've got the air conditioner on or they've got the heat blowing or whatever it may be. Or just something average happens. Someone walks into the room as you're giving a recording. So the re recording is done in places where we just didn't get a good enough recording that we have to do it again. How often does it happen, it really depends on the recording context in which someone is recording. So as long as you're in a quiet room, and have a headset microphone on, it should be fine, but you are going to get that once in awhile and will kind of recycle those sentences again so that we capture them. The sentences are ordered so that we're capturing the most important sentences first but that means if you make some errors we're going to have to recycle those again and get that. Again I want to stress that making a good sounding synthesized voice depends on the quality of the recordings that we initially get. Ken asks what is the format of these voice files. Can they be used with any free sharing device or just specific ones. The voices that we're building are compatible on all the different platforms. So all seven voices that we deployed in 2015 were on Windows platforms. And the voices work on Android as well and we're currently in the process of, our voices also work on the iOS platform, for iOS you have to integrate with each app and so that's what we're doing is we're starting off a handful of apps so that we don't haveone for all again this is just for our resource question at the moment. I think that if anyone has any other questions I'm happy to take them or please email us at hello at VocaliD dot C O. Thank you.