If the file contains interleaved PCM data with no header and you know the properties of the audio data (sample rate, number of channels, etc) the problem can be solved with basic math:
The number of bytes of audio data per second is sampleRate * bytesPerSample * numChannels
.
The starting offset in bytes would be then be bytesPerSecond * offsetInSeconds
, and the size of the chunk to read (in bytes) would be bytesPerSecond * lengthInSeconds
.