One of the language exercises that I learned at Middlebury - thanks to an instructor's great advice at one of the lunchtimes - was the shadowing technique. Essentially what it means is that you listen to an audio and speak along in real time - while recording yourself - trying to get your voice to match and predict the original reader's voice. It's hugely useful.
For Arabic, this is what I did:
- Use Al Jazeera's read aloud function (it also allows you to adjust the speed though I found the preset speed to be fine)
- Recite the text out in time with the voice
- Record yourself while reciting
- Listen to yourself. (I was completely horrified by how awful I sounded the first time, so bonus points if you don't cringe/cry.) Proceed to do this exercise again.
I actually found this to be really fun. You get to reaي an original text and while the computer generated voice is devoid of emotion it does manage to hit all the right أعراب. You can also do a similar exercise for English, as well as any other website that integrates a read aloud plugin or has transcripts available.
I also used this exercise to learn how to recite a poem for Middlebury's Arabic School poetry recital. I taped my teacher reciting the poem, and then recited the poem over her incredibly great voice for a few days, trying to capture the emotion and intonation. It's similar to how you'd lipsync a song - but this time you get to improve your language skills along with scoring a 100% in the 'How many lyrics of the *insert hopelessly embarrassing song that you were addicted to for weeks but would never admit to it* can you get right?' quiz.