Computer Algorithms Prefer Humans Over Other Voices

It has been revealed that deep learning systems, which are comprised of algorithms that can learn from other algorithms, prefer human voices, even though the systems do not actually read any text.

Scientists and robots have worked together to develop algorithms that can create predictions of whether a voice will sound positive or negative, like a natural voice or a computer generated voice. Using artificial intelligence systems, researchers developed a model that pairs a new human voice with an old voice for testing. Although the data was presented in the old voice, the algorithm still can pick up on aspects of the new voice that aren’t in the algorithm’s recognition chamber.

“It may be nice to describe humans as biased and unlikable, but it’s very difficult to program the world into computers that will generate similar judgments,” researcher David Palchak claimed.

We were curious as to what happens when they introduce new voices into the predictions for any given algorithm. Do the system’s responses change? Does the selection of new voices over time impact the predictions?

The research team was ultimately able to present two models. The first, which is called a whitelist, uses an algorithm to only choose an ideal computer voice. The algorithm chooses an optimal voice based on age, sex, and even nationality and geography. The second model has two other principles: “be like” and “distinctive”. In the whitelist model, an algorithm also selects a prototype voice that’s less diverse than the voice it started with, but for example, with a wider range of tone than the prototypical voice.

https://www.sciencedaily.co...

That method is useful in synthetic voice-producing. It allows a developer to find the most palatable and popular sounds. But not all artificial intelligence systems should be programmed with every sound. Those that do use the algorithms may use a list of possible sounds instead of one specific sound.