15.01.2013 Views

U. Glaeser

U. Glaeser

U. Glaeser

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

y a psychoacoustic audibility model, in order to meet the bit rate and the global threshold of hearing<br />

requirements. The whole procedure is described below.<br />

The psychoacoustic model filter bank is based on a 512-point FFT for layer I and on a 1024-point<br />

FFT for layers II and III. First, samples of the spectral power density P(k) and the respective level L P(k)<br />

in decibels are computed:<br />

© 2002 by CRC Press LLC<br />

P( k)<br />

X( k)<br />

2<br />

=<br />

LP( k)<br />

= 10 log P( k)<br />

(27.61a)<br />

(27.61b)<br />

where X(k) are DFT spectrum samples defined by Eq. (27.39a). Next in each of i = 0, 1,…,31 subbands<br />

the signal SPL is computed as<br />

Lpi = max[ LPmax i, 20 log10( 32768ssf max i)<br />

– 10]<br />

(27.62)<br />

where L Pmax i is the maximum L P(k) value in ith subband.<br />

Next, the relevant masker levels L Pm(z j) are searched for: tone masker levels L Ptm(z j), j = 1, 2,…,m tm,<br />

and noise masker levels L Pnm(z j), j = 1, 2,…,m nm. Then, the masking indices are computed (in dB):<br />

atm( zj) = – 1.525 – 0.275zj – 4.5<br />

anm( zj) = – 1.525 – 0.175zj – 0.5<br />

where by z j the jth critical band index in Bark is denoted.<br />

Individual masking threshold levels are computed (in dB) as<br />

Lttm( z, zj, LPm) = LPtm( zj) + atm( zj) + v( ∆z, zj) Ltnm( z, zj, LPm) = LPnm( zj) + anm( zj) + v( ∆z, zj) (27.63a)<br />

(27.63b)<br />

(27.64a)<br />

(27.64b)<br />

for tone maskers and for noise maskers, respectively. The so-called masking function v(∆z, z j) is defined by<br />

v( ∆z, zj) =<br />

⎧ 17( ∆z + 1)<br />

– 0.4L Pm( zj) + 6<br />

⎪<br />

⎪ ( 0.4L Pm( zj) + 6)∆z<br />

⎨<br />

⎪<br />

– 17∆z<br />

⎪<br />

⎩<br />

( ∆z – 1)<br />

( – 17 + 0.15LPm ( zj) ) + 17<br />

(27.65)<br />

where ∆z = z − z j. Expression (27.65) gives significant values in range −3 ≤ ∆z ≤ 8 only. Outside this<br />

region we can assume that v(∆z, z j) → −∞.<br />

Using expression (27.22), the global threshold of hearing L ptg (Fig. 27.14) can now be computed (in dB):<br />

where L ptq(z) is the threshold of audibility in quiet. Consequently,<br />

m tm<br />

(27.66a)<br />

(27.66b)<br />

for ith subband is computed over all critical bands j contained in this subband. Now in each subband<br />

the signal-to-mask ratio SMR i can be computed (in dB):<br />

10<br />

for<br />

for<br />

for<br />

for<br />

∆z < – 1<br />

– 1 ≤ ∆z < 0<br />

0 ≤ ∆z < 1<br />

1 ≤ ∆z<br />

Lptg( z)<br />

10 log10 10 Lptq z ( )/10<br />

10 Lttm z, zj , L ( Pm)/10<br />

∑<br />

10<br />

j=1<br />

Ltnm z, zj , L ⎛ ( Pm)/10⎞<br />

= ⎜ + + ∑<br />

⎟<br />

⎝ j=1<br />

⎠<br />

Lptg min() i = min( Lptg( zj) )<br />

SMRi =<br />

Lpi – Lptg min() i<br />

m nm<br />

(27.67)

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!