Most of the information here gathered for a analog project class
Scrambling audio usually refers to somehow garbling a sound signal being transmitted over some insecure medium (like a phone line, a walkie-talkie, cassette tape, etc) and ungarbling the sound at the receiving end, so that the intended recepient can understand the message. Typically, audio scrambling refers to mostly analog electronic techniques; purely digital setups like encryption are not considered scrambling.
Scrambling is intended as a low-cost solution for defeating a casual or a limited-resource eavesdropper. It won't stop someone truly determined from eventually figuring out what was said, but they can be slowed down for a while. A truly secure system would have to use aforementioned encryption techniques, which are expensive, requiring lots of computing power and circuitry to implement.
There are various scrambling methods that are commonly used, and commercial products are available for most of these. Their security tends to directly corrolate with their complexity and price; the usual tradeoff in these types of systems.
The various methods also have different bandwidth, processing delay, and quality tradeoffs, so a solution good for phone use (bandwidth ~ 5 kHz, low delay important for conversation) wouldn't neccessarily be good for scrambling done for cassette tape(bandwidth ~ 15 kHz, delay in scrambling unimportant).
Common methods include:
Any signal can be represented as a sum of its frequency components. (see Fourier Transform) With the proper circuitry, one can create a mirror image of an audio signal in frequency space. The designer picks some maximum frequency that they wish to allow (~3kHz works for acceptable voice quality,~20kHz if they want symphonic orchestra quality). Then, the slice of frequency space from 0 Hz to the maximum frequency is flipped around; the frequency components that were around the maximum in the original will be around 0 Hz in the scrambled signal, and vice versa. This is relatively easy to do with low-cost circuitry (the process is very similar to AM radio). But it isn't terribly secure; if an eavesdropper knows your magic frequency, they can easily descramble the signal. Also, people can actually train to understand such signals by ear.
Many designs of this nature are available for sale, of various levels of complexity. More sophisticated systems split the signal into two regions, and flip both regions separately. With even more effort, the splitting frequency can be varied based on some pattern that both the sender and receiver know, making it much harder for an eavesdropper to follow along. Some studies indicate that changing the band-division frequency roughly once per syllable results in the best scrambling security. All these additions, of course, increase cost. In general, this is probably the most common method out there. The scrambled signal does not take up any more bandwidth than the original, and quality can be quite good if enough care is taken in design.
In time rearrangement, blocks of the signal a few tenths of a second long are rearranged to a new order, and then transmitted. The receiver then reorders the sections to recover the original signal. As an example, say the sender's message consists sound block a, followed by blocks b,c,d, and e. The scrambler will store the blocks until enough have been entered, and then transmit them in a new order. Here, the output could be in the order d,b,a,c,e. With proper length of block, and sufficient blocks, any audio signal will be scrambled quite well. The main drawback to this method is the time delay the system invariably introduces, since it has to wait for enough of the message to be inputted before it can start output. Typical delays are close to a second, highly annoying for a real-time conversation. For recordings, the delay is immaterial. Also, it is difficult in purely analog systems to store blocks of signal for the required lengths of time. As such, this is not a very common system in use. However, it is fairly secure, and does not require any more bandwidth than the original message did.
With FM modulation scrambling, the method of creating FM radio signals is used, but instead of generating an output with a frequency in the MHz range, the scrambled signal is kept within the audio band (that is, less than 20 kilohertz). This system can have good quality(comparable, if not better, than FM radio), and can be very hard to unscramble, since an FM message looks nothing like the original signal, in time or frequency space. Companies have used this method for making secure cassette tapes, but beyond that I don't know of it being used. Due to the nature of FM transmission, the scrambled signal bandwidth can be traded off for power and quality. However, FM modulation is still more expensive than the frequency inversion schemes, and once someone knows you are using FM modulation, it is probably not very difficult to crack the message.
There are some other audio scrambling methods that I know of, and probably many more that I don't know of, but these cover much of the available devices, because the circuitry involved is relatively cheap, and the security is reasonable.
Now, if I could only put my own scrambler project together successfully