Comparison of four objective speech quality assessment methods based on international subjective evaluations of universal codecs

The performances of four objective speech quality assessment methods suggested to the International Telephone and Telegraph Consultative Committee (CCITT) by NTT/Japan, Bell Northern Research Canada (BNR/Canada), Centre National D´Etudes des Telecommunications (CNET/France), and US Department of Commerce, National Telecommunication Information Administration (NTIA/USA) are discussed. The four methods are compared on the basis of subjective opinion scores in the international measurement program for 16 kb/s codec standardization. It is shown that no method can provide objective scores agreeing with all subjective scores. If the method is allowed to be adjusted to each subject group, the NTT method can provide fairly good agreement. The BNR and NTIA methods need adjustment for 16 kb/s codecs as well as for each group. Significant improvement was not seen by combining multiple measures, as with the NTIA method