Some call it the boredom response script or the auto response script but what you're describing doesn't sound like that is the case.
The boredom script would have Hal talking in a question like format like:"Hello, is anyone there?". "Hey <username>, we were having a conversation weren't we?", etc. The script is usually called up after 20 to 30 seconds elapse without chatting to Hal.
If you heard various responses from your Hal in a conversational format then it could be that Hal was interpreting the server noise / fans as speech input and was responding to the gibberish as such.
This would require you to set your microphone adjustments and your speaker output to more suitable levels. You might try to reorient your microphone to a different location. I don't normally use a microphone but when I do, I use a headset mic with the noise cancelling foam at the end of the mouthpiece. Once my levels were adjusted Hal never made that mistake again. Here, in the states the headset can be had for about $24 USD, almost 30 euros.
A lot of users have found Hal and voice recognition in general somewhat inconsistent as far as quality and accuracy of interpretation regarding human speech. As a result, you might find yourself doing more editing of Hal's brain after using speech than is really worth your time.
With any speech program, the more you use it in a consistent manner with smooth, even, clearly spoken words and little to no background noise, the better your results.
Again, this is not a limitation of Hal but a limitation of speech recognition technology at present. It is getting better but it's not quite where a lot of us wish it to be.
Best of luck!