Although I agree with the general concerns raised in this thread FB is not sending raw audio to its servers and thus is not "listening" to you. The majority of

Although I agree with the general concerns raised in this thread FB is not sending raw audio to its servers and thus is not "listening" to you. The majority of work is done on the client-side. Specifically, a short audio sample is analysed client-side and turned into an audio fingerprint (the rough audio equivalent of hashing a value). That fingerprint alone is sent to the server for matching. Note that there is no way to reverse that process (hence the hash analogy) to get audible sounds out of it. Of course, this is all based on the assumption that Facebook is telling it like it is but given the fact that the difference in bandwidth use is an order of magnitude that's very easy to check.