Company : DeepVisionTech.ai
Sector : AI, Computer Vision, Accessibiity Products
My role : Product Strategy, User Research
Project Time : 2 Months
DeepVisionTech.ai is a stealth mode startup supported by Nvidia's Inception Program and Oracle for Startups. LetsTalkSign is a platform offered by DeepVisionTech.ai, it enables real-time sign language communication for the disabled. As of 2018, there are 466 million people with hearing impairment! Incorporating Accessibility at workplace is a challenge for the industries but a fundamental right for the disabled...
I did a user research on disabled people at workplace. DeepVisionTech.ai collaborated with Noida Deaf Society for conducting a survey of disabled people at workplace. This empathy-driven exercise was crucial to connect with the target audience, know about the challenges faced by the disabled people at the workplace. Sign Language Expert was consulted for the survey.
It should be noted that this project was accompanied during my internship tenure at DeepVisionTech.ai where I worked as a Machine Learning Engineer Intern. The study played a pivotal role in defining Data Augmentation Strategies, particularly with respect to the Indian Sign Language Context.
Research Goal:
#1 To identify target groups, their challenges at workplace and in general.
#2 Mode of communication
#3 Challenges with existing Assistive Technologies
#4 To strategize product framework
#5 To explore ways to satisfy Indian Deaf and Speech Impaired market
Competitive Analysis
Let'sTalkSign is a patented AI based device agnostic platform. No comparable solutions exist as of now. Hence, a thorough analysis of existing assistive technologies was done for its latency, price, ease of use. Alternative assistive technology includes Hearing Aids, Video Captioning, Audio based Transcription, etc. Apart from this, the AI algorithm was compared with other data-driven models for its resolution, memory specifications, algorithm training time, hardware requisites, latency.
"Transcription Apps are just okay!"
39.5% of the respondents felt that the existing assistive aids like Audio Transcriptions and Video Captioning performed Just Okay. And about 10 people considered their use to be "Not that much". Also, the audience had very little to no idea about the Let'sTalkSign so the following video was shown to them in order to make them aware of the product and the experience brought by the same.
After watching the video, 30 out of the 38 participants acknowledged that they had not come across such app before and 33 respondents found the Let'sTalkSign solution better than the existing assistive technologies used by them, including Audio Transcription and Video Captioning aids.
Opportunities and Challenges with LetsTalkSign
Opportunities/User Acceptance:
To help educational institutions in teaching disabled students : 18 people approve that Let'sTalkSign features are best suited for educational needs.
Attending events : 47% of the target audience believe that Let'sTalkSign would help them attend events, conferences, meetings.
Emergency Aid : Let'sTalkSign will also help people visit Police Station, Hospital, Court if the need be - without any external aid.
Workplace : More than 50% respondents consider Let'sTalkSign as a powerful tool that can help them in being productive at workplace.
Technical Challenges :
Training time for GANs : From my conversation with the authors of Research Paper, the training time of the GAN based architecture takes 7 days to train on array of high performing GPUs. The training time also reflects the computational complexity and as a consequence the cost required to come up with such a solution.
High Resolution Videos : Other video generating GANs that were considered were inferior in terms of the video resolution.
Indian Sign Language Data : The recent journals, conference proceedings related to Accessibility suggest that there is no common sign language that is adopted by everyone. The use of Sign Language, even, within India involves local contextual information in the language style. Thus, the dataset for training is not readily available over online repositories.
Pricing for the Solution : Considering the demographics of people, it was observed that the disabled people are not paid well - this might be true for larger audience as well because of the high dropout rates among the disabled people. With low salaries, people might not afford cost of the product.
Surveys
With principles of Design Thinking and UX Life Cycle, I conducted semi-structured interviews with the disabled people. Noida Deaf Society was contacted for the same as they empower disabled people representing different demographics. The main objective of the survey was to know more about the disabled people
Goals
#1 To capture opinions from diverse group of disabled people.
#2 To empathize and know the psychographics of the target audience of Let'sTalkSign.
#3 To identify challenges faced by the disabled people?
#4 To explore modes of communications used by them?
#5 To perceive the general sense of awareness about the Assistive technologies available to them.
#6 Get a feedback on Let'sTalkSign on various metrics that they feel are important.
#7 The popularity of conventional sign language in Indian scenario with local people in their day-to-day lives.
Findings of the Survey
Disturbed Workplace Gender Ratio for the disabled people : 35 people identified as Male. The low representation of female reflect the bitter fact that the Training program is mostly attended by males.
Low payscale among disabled people : 75% (27 people) get paid 5001 to 10000 INR per month which is low considering the cost of living.
Communication Gap with people : 33 people admit that there is a communication gap with people. A majority of people also report that they find it difficult to work in their organization as people misunderstand them. They claim that the disability is preventing them to communicate with people in a "casual" way. This is what a 23 year old at Noida Deaf Society :
"...People do not understand that I have disability"
Another respondent, 25, at Noida Deaf Society :
"I cannot do my work properly because of my disability. Unable to ask for help."
The qualitative data clearly indicates that there is a lack of empathy at workplace towards disabled people and they are considered "unproductive".
They compromise on Outcomes : Unfortunately, there were about 25% people who are forced to compromise on their outcomes and they are not productive.
User Personas
Based on the variety of the responses, I formally define target groups that reflect the needs and hopes of disabled people. This will eventually help feature distribution and sprint sequence.
This was particularly exciting! I came across Microsign Products, an organization with 60% of the staff - "differently abled". A product like Let'sTalkSign is very essential for such organization. Our survey also includes deaf people from Microsign. A moulding machine operator at Microsign reported:
"I am willing to pay for Let'sTalkSign."
Another target group is the students at the Noida Deaf Society. They share a common goal - to learn Sign Language as fast as possible. Meet Priyanka, from Noida Deaf Society.
It should be noted that the Name, Photo are sensitive information and the one used above is fictional. This is very important to protect the privacy of the respondents.
Product Mind Map
Customer Experience Journey Map
Conclusions
Accessibility is a basic right for the disabled people. But, in Indian Scenario, there is very less awareness. There is a emergent need of making workplaces more accessible. Let'sTalkSign has the potential to change the way disabled people communicate. Time has come for the organizations to empower them with the required support. As stated earlier, as of 2018 there are 466 million deaf people. WHO estimates show that the number will be 630 million by 2030 and a whopping 900 million by 2050. As we continue to innovate, it is very crucial to start thinking about them.
Disabled people at the workplace tend to earn very less. This is due to their educational qualification. Research has shown that the high dropout rates in school-going disabled children is mainly due to the "cognitive load". This is a serious concern for a world that envisions inclusive future. Let'sTalkSign can provide helpful interventions by enabling Sign Language Teaching. This tool helps both the teachers and students to learn the Sign Language.
The existing assistive technologies are suboptimal and require lot of innovation. They lack accuracy, ease of use. With the advent of ubiquitous computing, the idea of SIgn Language Communication App makes sense. Not only does it makes sense, it is logical to focus on training different sign languages to contextualize the local factors, regional languages, etc.
Let'sTalkSign looks promising with its claim of helping the disabled. People have reasonable expectations from the product. Their hopes and aspirations are a reflection of their innate will to be independent. The survey respondents have made it clear that the product is here to stay. There are a plethora of use cases where this app will be handy.
Although, there is no single sign language that accomodates all Indians. But training GANs on Indian Sign Language still helps! This is also evident from the suggestions received from few of the disabled people.
Learnings
#1 It is a surprise to know that "Communication" is stil a prevailing issue, despite the surge in ubiquitous computing. The technical inequity is so obvious while studying disabled people. There is a lot of scope of further research.
#2 The impact of the product is not only measured by its novelty, but also ease of access. The pricing of the product is critical for the success of the product. A detailed study of the target audience, their demographics, psychographics is required.
#3 Visual Design gives more clarity to the developers. The following UX helps a lot to streamline the scope of the product in general. This insights are important while building a prototype. The prototype is the best blueprint for the developers!
#4 The state of the art research might often be useless to address immediate concerns involving local problems of the people. There are several GANs that are more accurate, produce very high resolution videos and are useful for style transfer. However, it is not the best algorithms that wins, it is the robust algorithm that has the last laugh. I was personally involved in the GAN training process for Style Transfer. But the end-user is only concerned with their pressing pain points. Detailed surveys and interviews are essential to empathize with the target audience.
Note : My tenure as an intern also included implementation of GAN Architectures for its relevancy to the target audience. However, the architectures, model parameters, and other details related to the training of the AI algorithm are not shared as I am bound by a non-disclosure agreement with DeepVisionTech.ai. Since, the product is a patented technology, I have limited priviledge to share the details. But, having said that, I believe, all the details shared here are sufficient for the UX Portfolio Report.
Project Image Courtesy: https://www.pinterest.com/pin/133067363961859340/
