text and binary data in same file

open it as binary and read the text part into a string (or strings), or an array of bytes, then read the rest of the data as required.

It sounds like all you need the text mode for is reading the header size. In which case, you can just read/write that as binary as well.
The only difference (that i am aware of) between the two modes is that text mode will read each line up to a newline char which is system dependant (i.e. \n on Unix bases systems, \r\n on Windows). But since you are storing blocks of data then that is not a concern.

"Text mode" versus "binary mode" for opening the file is a totally separate concept from "text I/O" (really "formatted" I/O) versus "binary I/O" (really "raw" I/O). It is perfectly OK to use operator>> on a stream opened with ios::binary set, or .read() on a stream opened without ios::binary set.

Files opened with ios::binary set are always "raw" in the sense that no linefeed translation will be done and no particular byte value will be interpreted as end-of-file. However, this does not appear to be a concern for you.

If there is whitespace between "first symbol" (which I'm assuming is a "word" of text) and "corresponding data block", you will want to consume that from the stream (with stream >> std::ws) before reading the block. Also, you'll need to know the size of the block somehow. :)

In general, it's usually easier to just use raw representations of the numbers in question, and four-character codes instead of your "symbols" to identify blocks.