Figure 6.10
LSP tracks for a violin note played on the open A string sampled at 16 kHz (top)
compared to a spectrogram of the same recording (bottom).
Another plot shows the note as it was played, analysed through a time evolution.
This is shown in Figure 6.10, where the upper graph displays the 36th-order LSP values
as they evolve over time (note that although it looks as if they may do, the lines do
not cross, since they are always monotonically ordered), for each 512-sample analysis
window. Below this, a spectrogram of the same recording is plotted, using a 256-sample
window size with 50% overlap. Note the LSP narrowness around the obvious 440 Hz
fundamental and harmonics shown in the spectrogram. At the end of the played note,
some resonance continues to sound a 440 Hz fundamental but with only weak harmonics
(once the abrasion of the rosin-coated horsehair bow on the aluminium-clad string has
ceased). During this decay period the upper LSPs gradually lose their tendency to pair
up and begin to flatten out. However the fundamental continues to be marked by an
ongoing narrow spacing of the lower two lines.
The close visual correspondence between the spectrogram and the LSP evolution plot
in Figure 6.10 supports the assertion of Krishna and Sreenivas [9] mentioned previously
that LSP values are usable for instrument recognition. Most likely both fundamental and
harmonic frequency identification can be determined through automatic analysis of LSP
data, especially when combined with other measures such as ZCR or frame power.
154
Audio analysis
50
100
150
200
250
0
1000
2000
3000
4000
5000
6000
7000
8000
Analysis frame
Frequency
Time
Frequency
0.5
1
1.5
2
2.5
3
3.5
4
4.5
0
1000
2000
3000
4000
5000
6000
7000
8000
Do'stlaringiz bilan baham: |