Quality enhancement of CELP coded speech by using a voicing Gaussian mixture model

This paper presents a procedure to improve the quality of narrowband (0-4 kHz) CELP coded speech. The procedure is based on refining the pitch periodicity and reasserting the high frequency components (4-8 kHz) in the narrowband CELP decoded speech. The narrowband CELP decoded speech is first analyzed with a harmonic+noise analyzer and lowband information is extracted. By exploiting the lowband spectrum envelope V/UV information, the highband (4-8 kHz) spectrum envelope is recovered statistically by using a voiced/unvoiced Gaussian mixture model with interpolation. Lowband information along with the estimated highband information is then fed to the harmonic+noise synthesizer to re-synthesize wideband speech. Objective and subjective tests are performed to evaluate the quality of the re-synthesised wideband (0-8 kHz) speech. The results of experiments show that the re-synthesised wideband speech is pleasant to listen to with crispy characteristics and preferred over CELP coded speech.