BACKGROUND

The mixed excitation linear predictive (MELP) algorithm was adapted in 1996 as the federal standard for the 2.4-kbps voice compression algorithm. This algorithm provides intelligible speech provided the channel noise is low. However, military communications systems often suffer from high channel noise. The resulting bit errors adversely affect the speech intelligibility. To overcome this limitation, forward error correction (FEC) techniques are selectively applied to the transmitted bits (speech parameters), in order to optimize robustness of speech intelligibility in military communication. The ITT Aerospace/Communications Division applied a parameter design to determine the sensitivity of the MELP speech parameters, and it used the results for the design of an FEC that best protects the speech parameters from channel errors.

PROCESS

The ideal function was defined by taking the magnitude of input speech as the input signal and the magnitude of synthesized speech (output) as the response. Noise factors were speaker sex, speech phrases, acoustic noise and bit error rate. Design parameters taken as control factors were jitter, gain indices, bandpass voicing, pitch index and others. Control factors were assigned to an L ₃₆ orthogonal array. The signal-to-noise ratio was used to evaluate the process, which was thus optimized for robustness of speech intelligibility.

BENEFITS

• The optimum design resulted in a 3-dB improvement in S/N ratio, which is equivalent to reducing the effective bit-error rate by a factor of two.

• The results were obtained in a time severalfold...