Sign in
Search speechcoding
Subscribe to speechcoding
speechcoding by Keywords
ACELP | ADPCM | AMBE | AMR | AMR-NB | CELP | Codebook | DTMF | G.723 | G.726 | G.729 | GSM | Interpolation | LPC | LSF | LSP | MELP | PCM | Perceptual | Pitch | PSOL | QCELP | Quantization | SMV | VAD | Vocoder
Ads
Discussion Groups
Discussion Groups | Speech Coding | re: LPC filter order
Technical discussions related to Speech Coding (all itu and other vocoders, ACELP, CELP, AMR, etc)
LPC filter order - Satheesh Ram - Jan 19 8:45:00 2006
Hi all,
In vocoders, why linear prediction filter is always designed as 10th order?
From where does the magic number 10 came from? This question is cross posted from dsperado@dspe...
http://groups.google.com/group/dsperado/browse_frm/thread/6c2850f0ba4bd684/721849ed96f43992#721 849ed96f43992
Satheesh Ram
(You need to be a member of speechcoding -- send a blank email to speechcoding-subscribe@yahoogroups.com )
re: LPC filter order - Bill Wiese - Jan 19 19:58:00 2006
Satheesh..
> [Satheesh Ram <satheesh.ram@sath...]
> Subject: LPC filter order
>
> Hi all,
> In vocoders, why linear prediction filter is always
> designed as 10th order?
> From where does the magic number 10 came from?
Well, it's a 'rule of thumb':
number_of_LPC_terms = 4*bandwidth_in_KHz + 2
Why? Most (or at least many) voice codecs work with audio sampled at 8KHz (meaning 4kHz
bandwidth, but for practical telephony use, this is often restricted to 300-3.3kHz passband).
Now, for voiced speech, there's usually one formant peak per kilohertz of bandwidth.
It turns out you need two predictor terms to describe each formant peak - so there's 8 terms right
there. (I believe the original GSM vocoder used 8 LP terms.) Using an extra two extra LPC
coefficients further minimizes residual energy...
Bill Wiese
San Jose, CA USA
(You need to be a member of speechcoding -- send a blank email to speechcoding-subscribe@yahoogroups.com )
> [Satheesh Ram <satheesh.ram@sath...]
> Subject: LPC filter order
>
> Hi all,
> In vocoders, why linear prediction filter is always
> designed as 10th order?
> From where does the magic number 10 came from?
Well, it's a 'rule of thumb':
number_of_LPC_terms = 4*bandwidth_in_KHz + 2
Why? Most (or at least many) voice codecs work with audio sampled at 8KHz (meaning 4kHz
bandwidth, but for practical telephony use, this is often restricted to 300-3.3kHz passband).
Now, for voiced speech, there's usually one formant peak per kilohertz of bandwidth.
It turns out you need two predictor terms to describe each formant peak - so there's 8 terms right
there. (I believe the original GSM vocoder used 8 LP terms.) Using an extra two extra LPC
coefficients further minimizes residual energy...
Bill Wiese
San Jose, CA USA
(You need to be a member of speechcoding -- send a blank email to speechcoding-subscribe@yahoogroups.com )