This was my guess as well. The most common way to build up that much traffic is with an STP loop.

Typically CBWFQ (Class-based Weighted Fair Queueing) is not used for voice as it gives a bandwdith guarantee, but not one for latency. To keep voice quality consistent you want a fixed amount of bandwidth and the lowest latency possible. On cisco gear this is implemented using LLQ (Low Latency Queuing) which works as a straight priority queue. You can use policing to keep it from running away with all the bandwidth.