> On 12-Oct-99 Gabriel Bouvigne wrote:
> > I just wanted to notify that the voice option now only works (and is
only
> > tuned for) with 44.1kHz samples.
>
> Gabriel, could you post a quick rundown on how you implemented a 'voice'
option.
> voice is in the range 2-5khz isn't it? Could you assign zero bits to all
other
> freqs and still get reasonable sound? or would the harmonics wreck it?
>
> just wondering
> mike
The voice mode is made using 3 tricks:
*using only long blocks
*limiting bitrate when vbr
*using a band-pass filter
I tuned this mode using VBR (the old code), trying to reduce the average
bitrate keeping the same quality, wich to my mind leads to a quality
increase when in cbr. I used 2 voice sample: an english one, and the german
male voice from the squam cd.
Using short blocks helps to reduce pre-echo, but also reduces the quality of
this block. As there are no sharp attacks in speech, using only long blocks
helps to not reducing quality.
I limited the vbr bitrate to 160kbps, as I don't think that higher bitrates
are really usefull for speech only.
The last part is the band-pass filter. I've been very surprised with this
part. Theorically, voice can go from 800Hz to 3500Hz, in extreme cases. I
was thinking that it was possible to assign zeros to other freq.
Practically, it's not the case. Removing too much high freq leads to a sound
which seems more dull, but is still listenable. I choose to try keeping the
same sound as without the voice option, so I didn't go down to 3500Hz, and
tuned it using ears.
For the low freq, you can remove only a few. Otherwise, it leads to a kind
of metallic echo distortion.
As it was tuned using ears, I don't know exactly wich frequencies range is
kept. I'll have to compute it to know it.
Regards,
Gabriel Bouvigne - France
[EMAIL PROTECTED]
icq: 12138873
MP3' Tech: www.mp3tech.org
--
MP3 ENCODER mailing list ( http://geek.rcc.se/mp3encoder/ )
