The proposed algorithm requires firstly estimating the pitch period using the autocorrelation method. To this end it
is necessary to shift 2N samples, where N is the block size, 3 maximum values, 2 clipping operations and an N point
autocorrelation function. Once the pitch period has been estimated, the proposed algorithm computes Eqs. (1)–(3) together
with the value of factor_gen before to performing the time expansion of speech signal. Thus using the DSP TMS320C6711
and assuming that N = 80 samples with a sampling frequency of 8 kHz the proposed algorithm requires 2,033,100 cycles/s.
Finally taking in account that the DSP TMS320C6711 performs 8 instructions per cycle, the computational complexity of proposed algorithm is about 16.3 MIPS or about 8.2 MFLOPS. Table 2 summarizes the computational complexity of proposed algorithm. From this result it follows that proposed algorithm requires estimating only one autocorrelation function per block, which is a computational complexity lower of at least similar than that presented by the WSOLA algorithms that use the autocorrelation function as a similarity measure to determine the signal segment that must be used to reconstruct the time expanded signal, and then in general requires more than one correlation function per block.