Academia.eduAcademia.edu

Figure 7 – uploaded by Takehiro Moriya

See full PDFdownloadDownload figure
Fig. 7. MUSHRA difference scores (“with” minus “with- out” harmonic model) for MDCT based TCX.

Figure 7 MUSHRA difference scores (“with” minus “with- out” harmonic model) for MDCT based TCX.

Related Figures (7)

Fig. 1. Harmonic models (shaded arrows) either for con- text based or envelope based arithmetic coding. The proposed harmonic model is used to make the both types of arithmetic coding more efficient for tonal signals. Note that the perceptual noise shaping is carried out by a smoothed LPC envelope and does not take into account any harmonic model. On the other hand, the cod- ing error is shaped by a harmonic post-filter at the end of the decoding process. The post-filter is controlled by pa- rameters derived from a LTP analysis of the input signal.
Fig. 1. Harmonic models (shaded arrows) either for con- text based or envelope based arithmetic coding. The proposed harmonic model is used to make the both types of arithmetic coding more efficient for tonal signals. Note that the perceptual noise shaping is carried out by a smoothed LPC envelope and does not take into account any harmonic model. On the other hand, the cod- ing error is shaped by a harmonic post-filter at the end of the decoding process. The post-filter is controlled by pa- rameters derived from a LTP analysis of the input signal.
Fig. 2. Encoding parameters for harmonic model.
Fig. 2. Encoding parameters for harmonic model.
Table 1. Unequal resolution for coding harmonic intervals If the interval does not rely on the pitch lag of LTP, a hierarchical search is used to reduce the computational cost. If the index of the interval is less than 80, periodicity is checked by a coarse step of 4. After obtaining the opti- mum interval, a finer periodicity is searched around the optimum interval from -2 to +2.
Table 1. Unequal resolution for coding harmonic intervals If the interval does not rely on the pitch lag of LTP, a hierarchical search is used to reduce the computational cost. If the index of the interval is less than 80, periodicity is checked by a coarse step of 4. After obtaining the opti- mum interval, a finer periodicity is searched around the optimum interval from -2 to +2.
Fig. 3. Example of harmonic model exploited in a past context of the entropy coding. the conventional context is kept. The principle is depicted in Fig. 3.
Fig. 3. Example of harmonic model exploited in a past context of the entropy coding. the conventional context is kept. The principle is depicted in Fig. 3.
Fig. 4. Example of harmonic envelope combined with LPC envelope used in envelope based arithmetic coding
Fig. 4. Example of harmonic envelope combined with LPC envelope used in envelope based arithmetic coding
Fig. 6. Absolute MUSHRA scores comparing AMR-WB (15.85 kbps) and MDCT based TCX (9.6 kbps) “with” and “without” harmonic model. Number of listeners is 7. Fig. 5. Absolute POLQA scores of MDCT based TCX (left panel), and item by item difference scores (right panel, “with” minus “without” harmonic model)
Fig. 6. Absolute MUSHRA scores comparing AMR-WB (15.85 kbps) and MDCT based TCX (9.6 kbps) “with” and “without” harmonic model. Number of listeners is 7. Fig. 5. Absolute POLQA scores of MDCT based TCX (left panel), and item by item difference scores (right panel, “with” minus “without” harmonic model)
In Fig. 7, the same MUSHRA scores are used, but the item-by-item difference scores with and without the har- monic model are averaged over all 7 listeners. This time, we can see statistically significant advantages (95% double-sided student’s t-test) of the harmonic model for some music signals (two classics and one modern items) and on average. We observe that the proposed harmonic model doesn’t show any merit for the 5™ item (classic music) from the left. We suspect that this is due to its very complex harmonic structure.
In Fig. 7, the same MUSHRA scores are used, but the item-by-item difference scores with and without the har- monic model are averaged over all 7 listeners. This time, we can see statistically significant advantages (95% double-sided student’s t-test) of the harmonic model for some music signals (two classics and one modern items) and on average. We observe that the proposed harmonic model doesn’t show any merit for the 5™ item (classic music) from the left. We suspect that this is due to its very complex harmonic structure.
Related topics:
MathematicsComputer ScienceSpeech Coding
580 California St., Suite 400
San Francisco, CA, 94104
© 2025 Academia. All rights reserved