Tom Hanks never offered me a job. But it’s almost like he did.

Tom Hanks called me to get me a role in his movie, and he sure looks it.

Since PCWorld began covering the rise in internet usage, Various applications of artificial intelligence such as art of artificial intelligenceI’ve been looking through code repositories on GitHub and Reddit links, where people will post tweaks for their AI models for different approaches.

Some of these models end up on commercial websites. They either create their own algorithms or modify existing open source ones. An AI audio website is a great example. Uberduck.aiThere are hundreds of pre-programmed templates available at. You can enter text in the text box to have Elon Musk and Bill Gates, Peggy Hill or Alex Trebek, Beavis or The Joker virtually reading pre-programmed lines.

We uploaded a fake Bill Clinton photograph praising PCWorld in the last year. The model actually looks pretty good.

Training the AI ​​to reproduce speech involves uploading clear voice samples. The AI ​​learns how the speaker combines sounds with the goal of learning those relationships, perfecting them, and imitating the results. You may be familiar with the 1992 action thriller. Sport shoesYou are familiar with the scene where characters have to “crack” a biometric voice password. They do this by recording an audio of the voice of a target. It’s almost the exact same thing.

It can be difficult to build a vocal model that is good. There are often long samples to show how a person should sound. In the last few days, however, there has been a new trend: Microsoft Vall-E, Research PaperFor synthesized audio (with live examples), it takes only a few seconds to create fully-programmable audio.

AI researchers and other AI groups wanted information on whether the Vall-E model was yet available to the public. Although you can play with the Turtle model, it is not yet available for public use. (The Turtle is named after its slow speed, which it is, according to the author. However, it works.

Use the turtle to train your AI voice

The turtle is unique because you can train it on any sound you want by uploading just a few sound clips. The Turtle page on GitHubIt indicates that you should have several clips of approximately a dozen seconds. It will be saved as a WAV file with a certain quality.

How does it all work? You may not be aware that a public facility exists: google colab. Collab, which is a Google cloud service that gives access to a Python Server, is basically what it sounds like. You can store the code you write (or any other person) in a notebook that can be shared with others who have a Google account. The Turtle common resource here.

Although the interface is intimidating, it’s not too bad. First, log in as Google user. Next, click on Connect in top right corner. Be aware. Colab does NOT download anything to your Google Drive. However, other Colabs may. (The generated audio files can be downloaded to your computer, but they are stored in the browser. You may be running code written by another person. Bad input or a glitch in Google’s back-end could cause error messages. It’s all just a little experimental.

Each piece of code contains a small “play” icon that you can hover over. To run each block of code, you will need to click on it and wait for it to execute before moving on to the next.

Although we won’t be able to walk you through all of the features, please note that the red text can be modified by the user, such as the suggested text you would like the form text to read. You’ll find the option to train the form about seven blocks away. Name the model and then upload the audio files. Once you are done, you can define the sound model in the fourth block. Next, run the code and configure the text in third block. He runs That code block.

If everything goes according the plan, you will get a small audio output based on your voice sample. Does it work? I gave a quick and dirty sound sample to Gordon Mah Ong, whose work appears here. The Nerd PodcastThere are also many videos. I uploaded a sample that is several minutes long, instead of the snippet. Just to test it.

Results? This is it! SoundsLifelike, but not Gordon’s. It is currently safe from digital impersonation. This is not an endorsement for any fast food chain.

The turtle’s current model, which was trained on actor Tom Hanks, looks quite good. This is not Tom Hanks speaking here! As did Tom NotHe offered me a job but it was enough for at least one of my friends to believe.

conclusion? It’s a bit scary: The era of believing everything we hear (and soon will see) is over. Or, you already have.