Sexual coloring of the AI voice

In his example, the robotic vacuum cleaner formed an emotional attachment in young grandmothers. At 50, a woman's body undergoes a transformation, and there arises a need to care for a grandchild – a small helpless being. This need is partially fulfilled by the robotic vacuum cleaner.

In my seminars, I often ask the question: “In 30 years, a young single wealthy man will come to buy a home robot in a store where all possible robot images will be presented – from a cat to a grandmother. What appearance of the robot will the single man buy more than others?” Most often, listeners respond that it will be the images of beautiful girls.

At the same time, it is obvious that the pool of home robots cannot consist only of beautiful girls who communicate in such a way as to make their owner fall in love with them. For most young men, a beautiful woman nearby is important, but far from the only thing in his life. Among home robots, there will be many men, grandmothers, kittens, puppies, and virtual characters.

If the robot – a beautiful girl – only speaks in an erotic voice mode, the owner will quickly get bored, and emotional attachment will not form. It is important to alternate positive and negative emotions in the voice. Emotions of respect, admiration, and recognition of the owner's value are important. Emotional closeness and distancing from the owner are important, in particular, through alternating different voice modes.

Many girls currently give nicknames to their cars, talk to them, get angry at their breakdowns, and cry when selling their “beloved swallow.” In the navigator, girls set the most attractive male voice they can find.

Very soon, the car will start communicating with the owner in a voice that emotionally binds him to it. Probably, around the same time, AI assistants with the same function will become widespread.

About a year ago, publications emerged stating that ChatGPT was being used in New York for "phone sex" services. However, after censorship was introduced in ChatGPT, service providers began receiving outraged messages: "Your Mari understood me so well, but she has fallen out of love with me. Do something, bring her love back to me."

I don't know how much AI capabilities have developed in this segment at the moment. It would be interesting to hear from those in the know how advanced the technologies have become.

In this article, I would like to discuss one of the steps in the direction indicated by Igor Ashmanov.

Many mechanisms are involved in forming emotional attachment between a man and a woman. Many are tied to the exchange of emotional messages through voice. There are more complex aspects, such as: attachment style, adult-child-parent model, emotional swings – closer-further, confidence, etc., which I will explore in other articles.

▍ The Basis of the Attractiveness of the Female Voice

I will present data from two experiments.

1. In a sample of men, recordings of many voices were provided, and they were asked to choose the most beautiful among them. In fact, it was the voice of one girl at different stages of her hormonal cycle. All the men identified the most beautiful voice as the one recorded during the girl's ovulation period.

2. In the second experiment, an analysis of dialogues between a girl and different men was conducted. If the man was attractive to her, the girl's voice became higher.

There are many other signals that a girl colors her voice with when addressing a man she likes, but we will focus on pitch – it is easier to measure.

Conclusion: the bot should use various voice modes, sometimes including those signals that a girl unconsciously sends during ovulation to a man she finds attractive.

Currently, all the voices used by AI developers are very beautiful and attractive. Cartoonish voices are also used. AI voices are beginning to convey a lot of emotions.

But so far I have not heard of a bot giving a tone similar to "Je t’aime… moi non plus" by Gainsbourg and Birkin – a song that was banned but remained at the top in the late 60s.

▍ The question arises: how to make the voice as attractive as possible?

To train the neural network, data is needed. Ideally – the voice of a girl during ovulation and communication with a man she likes.

There are several approaches to collecting training data. The first approach is straightforward: talk to a thousand female students, select 1% of the most beautiful voices. Ask each owner of a beautiful voice to record dialogues with her favorite man on the day of ovulation. Experts will choose the most beautiful datasets to train the neural network that will color the AI's voice.

The second approach is to use ready-made data. They can be parsed from websites, or partnerships can be arranged with a site. Below is the approach I want to discuss with the experts at tekkix. I would appreciate any constructive criticism.

I am not very familiar with sources where good datasets for this task can be obtained. One hypothesis is the OnlyFans service. OnlyFans has 200 million male users and significantly fewer models. The site has many dialogues of an erotic nature.

There may be more accessible databases – dating sites, webcam sites, etc. Here, the opinion of experts is interesting. Let's consider OnlyFans as a source.

We can take recordings of the voices of models who are in the top 1% by revenue or other parameters. I believe that since they are at the top, they have attractive voices. We can manually check and remove unsuitable samples from this selection.

Try to calculate the days when the model is ovulating. Psychologists say that at this moment, the models have the highest revenue. After that, it is necessary to extract those datasets where the model's voice becomes higher (the man is liked).

Next, we will train a neural network on this dataset to distinguish the voices of maximum attractiveness from all other cases. Then we will train a neural network that will speak in the most pleasant voice, using feedback from the first neural network.

Regardless of the approach to selecting training data, a dataset of dialogues between women and men they find attractive will likely be collected, which can be used to train the neural network for emotional dialogues. Emotional scenarios (scenarios that respond to a partner's emotions) are an important element in forming emotional attachment. I will elaborate on this topic in more detail in the following articles.

Crowds of fans, enchanted by the magical voice, followed Ivan Kozlovsky, Muslim Magomaev, and Alla Pugacheva. I believe that in a few years, we will be surrounded by even more beautiful voices of AI assistants in our daily lives. And robots in adult services will surpass most experienced workers.