Comparison of Vocal Tract Length Normalization technique applied for clean and noisy speech

Sinusoidal-Based Lowband Synthesis for Artificial Speech Bandwidth Extension

“…practically unchanged. 4 Adding the proposed LB-ABE (approach LBUB-ABE) improves the AMR condition by −5.15 dB in LB-LSD, which corresponds to an improvement of −5.15 dB/(300 − 0) Hz • 100 = −1.72 dB per 100 Hz, thus exhibiting more than 4 times the restoring capabilities as compared to the UB-ABE approach.…”

“…In summary, test condition 4 shows that each speaker, male and female, takes profit from an LB extension. On average over speakers, LB extension improves the speech quality by a significant 0.26 CMOS points.…”

“…This observation may be explained by the fact, that the LB extension contributes to restoring the spectral balance of female speakers, who might suffer from an imbalanced spectral distribution after only applying UB-ABE. To understand this, we have to consider that female speakers in general have about 20% higher formant frequencies [3], [4], i.e., a spectral centroid on average at a higher frequency than men. Most of the formant frequencies are still present in the AMR condition, however, with a missing LB, a spectral imbalance towards high frequencies results, which consequently affects female speakers stronger than male speakers.…”

“…Together with the fact that on average woman have a 2-3 cm shorter vocal tract and consequently formant frequencies are approx. 20% higher than those of men [3], [4], leads to the observation that (on average) men have deeper voices than women.…”