Member ps_decoder_t::ps_get_in_speech (ps_decoder_t *ps): This function is retained for compatibility but should not be considered a reliable voice activity detector. It will always return 1 between calls to ps_start_utt() and ps_end_utt(). You probably want ps_endpointer_t, but for single frames of data you can also use ps_vad_t.
Member ps_decoder_t::ps_start_stream (ps_decoder_t *ps): This function is retained for compatibility, but its only effect is to reset the noise removal statistics, which are otherwise retained across utterances. You do not need to call it.