The CAMRAS format consists of sequences of 32 bits integers. The first bit is a sequence number, which should increment by 1 each frame. Then follow 255 integers for each channel.
The channels are 35MHz/256 ~= 136.7 kHz wide. Usually, the center frequency is at 420 MHz. In that case, the highest channel is at 441.6684 MHz.
The sampling time is 468.11429 µs.
The filterbank format is described here. It can be used with e.g. Presto. The header of the filterbank file is stored separately, because it contains useful metadata such as RA/DEC, time, Azimuth/elevation. The telescope is set to Effelsberg, because the filterbank format does not know the Dwingeloo telescope.
The archive format can be used with psrchive or the the Sigproc suite. If you have Docker, you can use these tools by running docker run -it caseyjlaw/pulsar-stack.
The numpy format can be read with NumPy. It contains one array, of dimensions (number of 10-second integrations, number of polarizations = 1, number of channels = 256, number of bins per pulse phase). The data has been dedispersed already. Metadata such as the dedispersion and period used are stored in the text file that also contains the filterbank header. Note that the channels are saved in decreasing order. The correct metadata for a center frequency of 420 MHz is: frequencies = 420. + 21.668359375 - 35.0*np.arange(256)/256.0.