Another possibility would be to use a gibberish language like
Simlish that is used in The Sims. I'm not sure how Simlish is implemented (they may have actually recorded whole sentences), but I imagine that with a reasonable number of word-like sound clips that can be joined together seamlessly in a random order, you could probably achieve a similar effect. Then you would just have variants that allow you to express making a question, anger, joy, surprise, etc. The subtitles would then be a "translation to English".