The accuracy of the speaker identification systems is degraded in the adverse acoustical environments by different kinds of interferences in the speech being input to the system. This corrupted speech contains portions, which are usable towards improvement in speaker identification. Usable speech is a novel concept of processing degraded speech data where the idea is to identify and extract these usable portions of degraded speech; as traditionally established by Spectral Autocorrelation Peak Valley Ratio (SAPVR) and Adjacent Peak Period Comparison (APPC). Unfortunately, the above-mentioned measures only detect about 74% usable speech with 26% false alarms. However, Usable speech has a harmonic structure whereas unusable speech has a noise-like structure. The information about the harmonics can be obtained using the ESPRIT (estimation of signal parameters via rotational invariance technique) short-time spectral envelope of the speech signal. The dominant peaks in the short-time spectra are the harmonic power for that spectral envelope. Consequently, the harmonic power originating from the periodic source is higher compared to the aperiodic unstructured source. This paper shows that this harmonic model-based usable speech approach has 82% success in identifying the usable speech frames with 18% false alarms.