Frank Klemm schrieb am Mon, 18 Sep 2000:
> :: > > The whole sfb21 thingy is a kludge, we should extend psymodel.c
> :: > > to calculate maskings for that band too.
> :: > >
> :: > the ATH is so large in that band, I would be afraid that
> :: > any computed maskings would always be < ATH, and thus
> :: > not worth computing?
> ::
> :: Wouldn't it be possible to use the ATH value as masking (as now), but use
> :: this value in the masking computing? It should prevent to lower a lot the
> :: overall masking for sfb21 when using something like --athlower 100. After
> :: all in the others sfb when using --athlower 100 the whole masking is not
> :: reduced, only the ATH. I'm afraid that now, using --athlower 100 would
> :: result in an incredible ammount of bits in the latest sfb. I'll test it and
> :: post the results.
> ::
> The main problem of masking at low SPLs is that is weaker than at higher
> SPLs. At 30 dB(A) low frequency and high frequency slope are nearly
> identical. Masking is relative to the masking signal higher, but much faster
> falling into high frequency direction.
>
> x 90 dB(A)
>
> ___
> \___ ||
> \___/ \
> \___ ||
> \___ ||
> x 55 dB(A) \___/ \
> _ \___
> \_ \___
> \_
> \_
> \_
> \_
> x\ 40 dB(A)\_
> \ \_
> \ \_
>
> --
> Mit freundlichen Gr��en
> Frank Klemm
Frank, that's not what Gaby is talking about.
But if you are talking about the spreading function, there
are more parameters than loudness:
- frequency
- tonality
- temporal effects
- difference tones reducing masking
Ciao Robert
--
MP3 ENCODER mailing list ( http://geek.rcc.se/mp3encoder/ )
