«I believe that a teenager should be able
to create
a virtual assistant in half
an hour». How platform developers are making speech
technologies accessible to all
The Report: Facts
and figures
Facts
and
figures
In 2018, Russian developer and entrepreneur Stanislav Ashmanov set out to launch SOVA — an Open Source platform for creating chatbots and voice assistants. A year later his project got a green light from The Project Support Fund of The National Technology Initiative (NTI). The Fund will award Stanislav more than 300 million rubles to develop the platform. Read more to find out how SOVA will work and why it is crucial to make technologies accessible.
Stanislav created his first virtual assistant at the age of 12 and taught it to solve simple maths problems. Interest in science and technology runs in the family: Stanislav’s grandfather worked in linear programming (a field of mathematics), while his father became interested in speech technologies at the Computing Center of the Academy of Sciences in the 1980s and was the first in Russia to develop automatic spell checking. As a child, Stanislav was interested in robots: he studied mechanics and programming, participated in student competitions and followed robotics contests held at the Lomonosov Moscow State University (MSU). Eventually, he became a student at the MSU’s faculty of mechanics and mathematics.
In 2015, a year after his graduation, Stanislav founded «Ashmanov Neural Networks». «I’ve always been interested in the synergy between man and machine. In my final years at university I realized that I should start a project in artificial intelligence. After launching the company, we tested a number of robot-related projects. For example, in 2013, we were busy with the launch of the „Lexi“ project and we were also developing a smart speaker with voice assistance. Eventually we started working with big data analytics: we received orders, created recommendation systems, built recognition algorithms for faces and car number plates, and so on», — Stanislav recalls. His clients included Mail.ru, NEC and other major companies. In 2017, Stanislav took over the management of his father’s company «Nanosemantics». Although a business merger hadn’t formally taken place, «Ashmanov Neural Networks» and «Nanosemantics» started working as a single entity.
The business kept growing. However, in 2018, Stanislav realized that he wanted to launch a non-profit project, which would benefit the IT community, and the idea of an Open Source platform was born. It would allow users to create a virtual text or voice assistant. «After researching the market, I realized that there were mostly open access kits for programmers, which could not be used by a non-professional. Whereas I believe that a teenager should be able to create a virtual assistant in half an hour», — says Stanislav.
Info
What is Open Source?
Open Source software, as opposed to paid software, is accessible to all – in full or in part.
Such software is distributed under an open license, thus, anyone can use it. GitHub is the largest archive of Open Source solutions in the world. Independent developers as well as large companies such as Facebook, Apple, Google and others upload their repositories (code archives) there. The IT community views Open Source as one of the key drivers of technological development in recent years.
Stanislav confirms that creating projects in speech technologies is not an easy task. The essential components of a voice assistant include speech recognition (the virtual assistant’s «hearing») and speech synthesis (its «voice»). Before teaching a virtual assistant how to understand human speech and «talk», you need to amass data (a dataset) for training neural networks and to mark it up. Datasets with voice recordings are freely available in the public domain, however, they might be of poor quality or unsuitable for the developers’ requirements. One would also need to pay to have large volumes of data marked up. While corporations and banks can afford to hire professional speakers to record high-quality datasets and then pay for marking them up, the start-ups and independent developers cannot.
He named his new project SOVA (Smart Open Virtual Assistant) and engaged part of his team in its implementation. However, both «Ashmanov Neural Networks» and «Nanosemantics» were still mainly engaged in commercial work. At this point Ashmanov started looking for funding. «I knew that the project would be worthwhile, if we were able to create a platform for users from around the world. Thus, besides Russian, we would have to work with other languages such as English, Chinese and Spanish. We worked on the budget and started looking for an investor. We considered a number of options — from crowdfunding platforms to venture funds. The funding eventually found us, and it happened quite quickly. It was sometime in 2018, when I was alone in the office late one night and heard a phone ringing in the commercial department. I picked it up and spoke to people from NTI and RVC — they were interested in SOVA. I knew about RVC, but didn’t think that they would be interested in our project. They explained that in order to receive funding, a project must provide a platform solution, which would benefit the entire market. I replied that our product does just that», — remembers Stanislav.
Negotiations lasted for almost a year, during which the project had to be significantly re-worked. «We got rid of some parts of the project, and added others. For example, Dmitry Peskov [CEO of the „NTI Platform“ NGO and a special representative of the President of the Russian Federation for digital and technological development] specified that the project must be implemented as a consortium — in other words, a team of developers should be contributing to our Open Source platform. So, for example, MIPT and FEFU joined our consortium — they now share audio recordings of their lectures with us. Functioning as a consortium gives more weight to the non-profit idea behind the project», — he says. In August 2019, the project received final approval from the Interagency Working Group (IWG), which is responsible for the implementation of the NTI. The NTI Project Support Fund, which has been created by RVC, decided to invest more than 300 million rubles in SOVA. «When we learned that our project was approved by the Fund, and that we will be receiving investments, we immediately started looking for new staff. At that point we realized that we had enough resources to launch the platform», — says Stanislav. «It’s not just money to spend. We have KPIs, and the Fund makes sure that they are achieved. Like any investor, the Fund wants to ensure that we use the money effectively to attain results. Hence, we receive funding in parts».
Details
8 billion
voice assistants will be in use globally by 2023, according to the forecast by Juniper Research.
This means that each active user will have two to three virtual assistants. At the same time, according to a study by CB Insights, the market for voice assistants grew to $49 billion in 2019. Corporations such as Apple, Google and Amazon are the main beneficiaries of this trend, as smaller companies and independent developers do not have enough resources to create comparable technologies.
At present, approximately 80 people are working on the development of SOVA platform — these include developers, data scientists, linguists, and other specialists. Stanislav highlights the fact that the project didn’t start from scratch, as the team had access to all previous work-in-progress created by «Ashmanov Neural Networks» and «Nanosemantics», which implement projects for the largest companies in Russia and the CIS, as well as for government departments. The team started working on launching the company’s products onto global markets. The first release is scheduled for August 2020: programmers from around the world will be able to download Open Source codes from GitHub. «Any programmer will be able to download everything that he or she needs. We will publish algorithms and data (high-quality marked-up voice recordings, texts, dictionaries, etc.), which can be used to train these algorithms. We will also provide development tools, so that users can add skills for chatbots or voice assistants», — explains Stanislav. Then these virtual assistants, whether text or voice, can be used in a number of ways — embedded in websites, mobile applications, or even in external devices (smart speakers, vehicles, etc.).
It’s important to give developers the ability to create virtual assistants to suit their needs, says Stanislav: «For example, someone would want to develop a virtual assistant that will speak about birds. We cannot foresee this, but we can allow this user to upload a reference book about birds and train the algorithm to answer relevant questions». Developers without personal datasets will be able to choose a «voice» for their assistant: SOVA will provide high-quality datasets with pre-recorded voices of men, women, and even children. Programmers will also be able to use ready-made dictionaries and scripts, which will help train virtual assistants to engage in dialogue, for example, greeting users in response to their greeting, and so on.
The launch of the SOVA Store — a marketplace for the community of developers — will be a logical next step. Users will be able to upload modules for creating and training virtual assistants onto the SOVA platform, so that other developers can use them (some modules will be free, some would need to be purchased). In essence, modules allow to expand the capabilities of virtual assistants created on the platform. Stanislav is certain that a complex SOVA ecosystem will gradually emerge, and the platform will grow thanks to its users. Initially, only programmers will be able to use the platform. However, developers are working to ensure that people without special training will also be able to access SOVA in the future: there will be no need to dig deep into the code to make amendments, as all functions will be accessible via a simple and intuitive interface. «We envision, that the majority of our users will be independent developers and small businesses. We are already working with large companies: our current customers include corporations, Telecom operators, and ministries. We will continue working with them, whereas the SOVA platform will meet the needs of those companies and individuals who cannot afford to pay for the development of expensive software», — Stanislav explains.
The team is also working on creating its own SOVA voice assistant and the accompanying hardware, but its key goal is to launch the platform. Stanislav is certain that the market for virtual assistants will be transformed in the future, and they will become more and more customized. Users will be able to create personal voice assistants and customize them by choosing a voice, character, sense of humor, and much more. Mass distribution is another trend: people around the world are getting accustomed to using their voices to communicate with their devices, and soon vending and washing machines with built-in voice assistants will become commonplace. On balance, Stanislav believes that virtual assistants will no longer be perceived as robots, and that’s the most important trend. «I think that people need voice assistants that are able to communicate with them, and not just perform simple functions like playing music or ordering pizza. Such communication is especially vital for older people, who might find smartphones too complicated to use, and for children. I’m sure that voice assistants will become much more humanlike».
Info
Projects supported by NTI in 2019:
«CableWalker» drone-based electric grid maintenance system
«Kursir» mobile complex for inspections of flight support radio equipment
«Sputnix» scalable platform for low-orbit small spacecraft
«AntionkoRAN-M» gene therapy antitumor pharmaceutical
Digital technologies used in the production of high-temperature superconductors (implemented by «S-Innovations»)
ERA cargo drone
Competitions in engineering and marine robotics
Russia’s first testing site for the certification of unmanned aircraft
Russian Unmanned aircraft system Traffic Management RUTM-1
«ꓯPlatform» distributed energy management
SOVA open platform for virtual assistants
Virry virtual reality for psychoprophylaxis
«Russia is a country of seafarers of the XXI century» project
