Note: this part of the exercise may require some basic coding skills (e.g., in Python). Students who cannot code will be given help in lab sessions to find an alternative.
You might select sentences from a (large) body of existing material, possibly using an algorithm of your own. Or you might generate the sentences algorithmically, either using your own method, or a Large Language Model. You could even combine these approaches (use an LLM to generate a large corpus, then select carefully from that).
Select enough sentences to yield around 30 minutes of speech.
Remember that you’ll need to record this material in the studio! Make sure to record under exactly the same conditions (same studio, same microphone, etc) in all of your recording sessions, so that you have the option of combining the data to train a single model.
Some ideas to get you started:
-
- Replicate the ARCTIC text selection algorithm, or try your own ideas (but keep it simple!)
- Use much more up-to-date source text, instead of old novels.
- Use a Large Language Model to generate text.