A PyCUDA and PP parallelized viterbi decoder. Pyterbi requires CUDA compute capability of 2.0 or greater. Pyterbi is about 100 times faster than straight python at decoding a Viterbi trellis of 64 states and 1024 observations. Pyterbi will provide a roughly linear speedup with the number of processor cores if asked. All code and documentation is under an MIT license.