On Jul 15, 12:07�am, Vladimir Vassilevsky <antispam_bo...@hotmail.com>
wrote:
> sooridas wrote:
> > 1. Voice/Unvoice Decision using speech samples
> > �What is the threshold for zero crossing rate of the speech signals for fs
> > =8000 with block length 240 so that i can safely assume that anything below
> > that can be flagged as voiced frame?
>
> The Voice/Unvoice distinction is very relative and rather artificial.
>
>
>
> > 2. Pitch Detection using auto correlation coefficients.
>
> > I took the auto corr coefficients from middle, search for zero crossing of
> > the auto corr coefficient after 24 samples. Then searched for the �the
> > maximum peak after the zero crossing of auto corr coefficients. The index
> > of this second max peak is � taken as the pitch if the energy of the �same
> > is greater than 0.30 times the R(0).
> > Is this can be reliabilty used for pitch detection?
>
> No, it can not. Care should be taken about the pitch multiplication problem.
>
> > 3. How do i get the the SNR and MOS of my basic LPC �codec. Is there any
> > help available in matlab for the same.
>
> The SNR doesn't make sense since your LPC is not a waveform coder. The
> MOS figure can be estimated numerically according to ITU-T P 8xx
> specifications.
I really don't trust opinionated maths. :-)
Steve
Reply by Vladimir Vassilevsky●July 14, 20082008-07-14
sooridas wrote:
> 1. Voice/Unvoice Decision using speech samples
> What is the threshold for zero crossing rate of the speech signals for fs
> =8000 with block length 240 so that i can safely assume that anything below
> that can be flagged as voiced frame?
The Voice/Unvoice distinction is very relative and rather artificial.
>
> 2. Pitch Detection using auto correlation coefficients.
>
> I took the auto corr coefficients from middle, search for zero crossing of
> the auto corr coefficient after 24 samples. Then searched for the the
> maximum peak after the zero crossing of auto corr coefficients. The index
> of this second max peak is taken as the pitch if the energy of the same
> is greater than 0.30 times the R(0).
> Is this can be reliabilty used for pitch detection?
No, it can not. Care should be taken about the pitch multiplication problem.
> 3. How do i get the the SNR and MOS of my basic LPC codec. Is there any
> help available in matlab for the same.
The SNR doesn't make sense since your LPC is not a waveform coder. The
MOS figure can be estimated numerically according to ITU-T P 8xx
specifications.
Vladimir Vassilevsky
DSP and Mixed Signal Design Consultant
http://www.abvolt.com
Reply by dbell●July 14, 20082008-07-14
On Jul 14, 8:54�am, "sooridas" <soori...@gmail.com> wrote:
> Hi Friends,
> � � � � � � I have implemented a basic LPC coder in matlab using the
> following blocks. Kindly let me know your comments on some of my questions
> detailed down.
>
> I have taken sample .wav file with
> sampling frequency -8000
> frame length =20 ms = 160 samples
> Overlap = 50%
> Block �length � = 240 sampels
>
> Overlapping samples of size 240 is copied from the input speech file. I
> windowed the samples with a hamming window.In order to cancel the effect of
> one pole in the glottal �response, i pre-emphasised the signal. The LPC
> filter coefficnets Ak are found out and this is inverse filtered to get he
> residue. Pitch and voicing decision and gain is found out from auto
> correlation coefficients using zero crossing. This information is used to
> decode the signal by synthesizing the source signal. Overlap and added the
> prev samples and then de-empasised the signal to get the synthesized
> output.
>
> I would like to know
>
> 1. Voice/Unvoice Decision using speech samples
> �What is the threshold for zero crossing rate of the speech signals for fs
> =8000 with block length 240 so that i can safely assume that anything below
> that can be flagged as voiced frame?
>
> 2. Pitch Detection using auto correlation coefficients.
>
> I took the auto corr coefficients from middle, search for zero crossing of
> the auto corr coefficient after 24 samples. Then searched for the �the
> maximum peak after the zero crossing of auto corr coefficients. The index
> of this second max peak is � taken as the pitch if the energy of the �same
> is greater than 0.30 times the R(0).
> Is this can be reliabilty used for pitch detection?
>
> 3. How do i get the the SNR and MOS of my basic LPC �codec. Is there any
> help available in matlab for the same.
>
> Regards,
> Sooraj.
What kind of LPC are trying to implement? What is the exitation? What
is the bit rate? Does calculating SNR make any sense for the version
you are implementing, i.e. is there any effort to duplicate waveforms
rather than the way it sounds?
Dirk
Reply by sooridas●July 14, 20082008-07-14
Hi Friends,
I have implemented a basic LPC coder in matlab using the
following blocks. Kindly let me know your comments on some of my questions
detailed down.
I have taken sample .wav file with
sampling frequency -8000
frame length =20 ms = 160 samples
Overlap = 50%
Block length = 240 sampels
Overlapping samples of size 240 is copied from the input speech file. I
windowed the samples with a hamming window.In order to cancel the effect of
one pole in the glottal response, i pre-emphasised the signal. The LPC
filter coefficnets Ak are found out and this is inverse filtered to get he
residue. Pitch and voicing decision and gain is found out from auto
correlation coefficients using zero crossing. This information is used to
decode the signal by synthesizing the source signal. Overlap and added the
prev samples and then de-empasised the signal to get the synthesized
output.
I would like to know
1. Voice/Unvoice Decision using speech samples
What is the threshold for zero crossing rate of the speech signals for fs
=8000 with block length 240 so that i can safely assume that anything below
that can be flagged as voiced frame?
2. Pitch Detection using auto correlation coefficients.
I took the auto corr coefficients from middle, search for zero crossing of
the auto corr coefficient after 24 samples. Then searched for the the
maximum peak after the zero crossing of auto corr coefficients. The index
of this second max peak is taken as the pitch if the energy of the same
is greater than 0.30 times the R(0).
Is this can be reliabilty used for pitch detection?
3. How do i get the the SNR and MOS of my basic LPC codec. Is there any
help available in matlab for the same.
Regards,
Sooraj.