Shulou(Shulou.com)06/02 Report--

In 2016, James Vlahos, an American science journalist, did something that moved countless people.

A few months before his father's death, he was determined to keep his father's voice and teachings with him forever. So James with a non-technical background taught himself speech synthesis and machine learning with the help of an AI project. With the help of his father, he recorded his father's voice for an hour or two every day, recording more than 90, 000 words to train the AI model. Finally completed a voice assistant made up of his father's voice, similar to siri, which made James mourn from time to time.

This story not only moved countless families around the world, but also made AI developers and technicians see the importance of customizing AI voice. There is no doubt that many families around the world are longing for similar functions, whether it is recording the voice of the elderly, so that their own voice can accompany more children to grow up, or the voice companionship between lovers, the family is becoming the main battlefield of AI voice technology application scenarios.

This demand is also being paid more and more attention by the industry. In recent years, speech synthesis, voice cloning and other technologies have been developed, and the overall ability of natural language processing has also developed by leaps and bounds. The customization of AI voice does not take months to carry out machine learning training with tens of thousands of corpus, but is really "now flying into the homes of ordinary people."

In early March, Baidu, which has been investing heavily in AI technology, launched a voice customization function in a small speaker. In the scenario of "Mom and Dad tell stories" in Xiaodu APP, users can record voice packages of themselves and their families.

This is the first time that user voice customization capability has appeared in conversational AI hardware. When users can customize their own voice packages and let smart speakers keep their voices coming, many industry rules seem to be changing.

From the confluence of speech synthesis, conversational AI, and intelligent voice hardware, let's take a look at the three possible changes that may take place in 2020 as we enter the era of AI voice customization.

The threshold has gone: AI Voice has entered the era of customization

In fact, the ability of AI voice customization has always been highly expected by the AI industry and users. On the one hand, let AI simulate the user's voice, related to family, companionship, memory and many other social and emotional factors; on the other hand, familiar sound may trigger a lot of new application imagination, for example, you may not bother to open audio lessons, but if your love bean or goddess gives you audio lessons, you may not even bother to sleep.

Therefore, the engineering and commercial application of AI voice customization has always been highly expected. This technical clue can be said to be a marvelous soldier in the continuous development of AI voice hardware, such as smart speakers, smart screens and other products.

The related technologies of AI voice customization have ushered in the process of lowering the threshold and increasing the application scale in the past few years. James Vlahos uses more than 90, 000 corpus for machine learning training, but now it only takes a few minutes to train semantic comprehension and natural language processing that are far better than siri's customized speech model.

In recent years, with the upgrading of technology, the exploration of customized user voice industry has been moving forward. For example, a public welfare project called Revoice hopes to help people with ALS retain their own voice, while car AI maker Cerence launched the function of creating a user's voice assistant last year, and Microsoft's Custom Voice service can make a user's voice into Xiao Bing's voice to some extent. Last year, "voice customization" began to be applied to map scenes, and users were able to generate a complete personal voice package by recording 20 sentences on Baidu Map APP.

Today, the function of customized voice has come to the most complex AI scenario: conversational AI devices.

In the Xiaodu voice customization feature, users can record their own voice packages in the "Mom and Dad Story" function when they enter Xiaodu APP. Needless to say, it can be recorded in 3-5 minutes, and the recorded voice can tell a long story, and the tone, tone and sudden frustration are very lifelike with Baidu's AI voice ability.

This means that there is basically no user threshold for AI's ability to customize voice, and we don't have to learn complex technologies, waste a lot of time, and endure failures again and again. Users will be able to use intelligent voice customization to implement applications in home scenarios in a very simple way. The industrialization channel of voice customization is also launched.

On another track, we can think of it as the overall evolution of intelligent voice assistant and conversational AI hardware.

Since Amazon's Ehco was born in 2015, voice assistants have been in the basic ability to ask and answer questions with machine tones. Users often can't find the motivation to go on. The question-and-answer model is not similar to human interaction.

In 2019, Xiao du's assistant realized the full-duplex non-wake-up ability, which could wake up multiple interactions at a time, and finally enabled multiple rounds of conversations to be realized in hardware, and the chat began to look like a real person.

The ability of AI voice customization may be regarded as another upgrade of intelligent voice assistant and related hardware in 2020, through which users can achieve thousands of AI hardware, and developers have a new development foundation. The industrialization impact of chain can also be carried out.

Food Circle & Family: AI hardware or burst in two scenarios

The first change brought about by AI voice customization is that users may start thinking again about how they use conversational AI hardware and why they buy it.

With AI voice customization capabilities, two business scenarios are obvious. First of all, in family scenarios, the ability to customize family voices is actually crucial. Because the voice of the family represents companionship, dependence and warmth, this is human nature and cannot be changed at any time. Use parents' voices to tell stories and knowledge to children, let children's voices accompany parents in smart speakers, tell parents time and read news. These warm applications are not only the general needs of the Chinese people, but also the inevitable choice for working in busy cities.

Today's situation is a case in point. The epidemic has delayed the resumption of work, which has given many parents more time to spend with their children, resulting in "parental dependence under the epidemic." But when the rework begins, what if parents have no choice but to leave their children? In the case of home use of smart speakers, the voice customization function gives an option.

On the other hand, the bigger dividend of AI voice customization depends on the food circle. The great energy of the rice circle these days has been quite appreciated by the whole society. So let Aidou's voice not only appear in map navigation, but stay in smart products all the time, talking to yourself, chatting, telling stories, and playing games-- the purchasing power and redevelopment power generated by this can hardly be thought about.

These two scenarios are most likely to break out quickly under the voice customization capability of AI. And on that basis, a new wave of developer bonuses is coming online.

Generalized customization: AI voice developers get new tickets

With the maturity of AI voice industry and the increasingly complete technical support of developers, more and more sound bloggers and AI developers have devoted themselves to the spring tide of AI voice ecology. With the launch of the AI voice customization function, the basic capabilities of developers have made a big breakthrough, and the conversational AI device of "thousands of people and thousands of voices" is no longer just an industrial imagination.

AI voice developers may soon be able to get a new opportunity for "generalization customization" through voice customization. It can be predicted that AI voice customization will affect the development space and industry value of AI voice in the following ways:

1. Skill customization has been developed rapidly. Customizing a voice skill with the voice of a family, or even a voice skill exclusive to family members, couples and fans, is a broad industry imagination. Many voice skills will change completely with the option of user voice, which may affect entertainment, family, education, companionship and so on.

2. Customization of life scenes has become an important play. Hearing the voices of your relatives and idols in smart homes, smart phone assistants and smart wearable devices is a thing that can be full of all kinds of games. Developers will be able to use a variety of hardware forms to wield the imagination of AI voice customization.

3. Numerous new ways to play "sound copyright". As mentioned above, the emergence and popularization of AI voice customization capabilities will make "high net worth voice" a new copyright capital. The voices of stars, idols, public figures, and even Internet celebrities in specific areas can be popularized to all kinds of hardware in the form of AI interaction, creating another vertical tuyere between the content industry and the technology industry.

AI voice pan-customized applications, hardware, and proprietary services, which can be implemented on a large scale, are a new form that combines users, idols, software developers and hardware brands, and the resulting purchase desire and platform development opportunities may be a unique landscape in 2020.

4. The social value and significance of AI voice have been re-evaluated. From the story of James Vlahos, it is not difficult to see that AI voice customization ability contains profound and meaningful family care and family significance. People cannot be accompanied forever, but the intelligence of each other's voice can magnify many important moments and companionship. Developers of AI voice customization are likely to take on more exploration of affection, society and companionship. From technological value to social value, the influence of AI voice customization will also be magnified.

AI voice customization is becoming a new driver in the conversational AI hardware market. If we carefully observe the conversational AI hardware and AI voice market in the past three years, we will find that the fluctuating growth of the market is closely related to technological breakthroughs. When a hardware form is in its infancy, this kind of commercial energy caused by technology is the norm of the industry.

In other words, the hardware market opened by conversational AI presents the logical relationship that technological capability breakthroughs represent a better user experience, which in turn will directly lead to market feedback. In 2019, after Xiaodu brought full-duplex non-wake-up capability, the AI voice hardware market once unsealed the tripod, showing a big leap forward alone. And AI voice customization capability, as a technological breakthrough closely related to developers, skill ecology and content ecology, will obviously continue to maintain this technology leadership and bring more market feedback, so that a certain market qualitative change is approaching.

But no matter which platform gets the final right of retention, for AI developers, the industrial opportunities brought about by voice customization are just beginning. The hardware with thousands of people, the ever-changing applications, and the technological breakthroughs by all means are the results that we finally want to see in the new hardware form.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.