I think there are an infinite number of AI possibilities and different use cases. As a general matter, we would advocate for a copyright exception that says if you have access to a work, you should be able to use a computer to analyze that work, compare it to other works and look for correlations and patterns, which can then be used to develop an AI model for future things. In your case, this would mean voice recognition. What you might need is a sort of large corpus of recorded speaking. In addition, you might also need transcripts of those recordings.
You would then train an AI system based on.... It would be a very large corpus with hundreds of thousands of hours of voice recordings and then the transcripts to create a model, so that the AI system looks for the patterns, matches the voice to the transcript, and can do it again in the future when it hears a new speech.