[hts-users:03601] Re: HTS-demo pulsenoise

[Keiichiro Oura <uratec@xxxxxxxxxxxx>]

Hi,

I don't know other frequency.
But, I guess that there is speaker-dependency.
For example, if you use husky voice, the frequency should be changed.

Regards,
Keiichiro Oura


2013/2/8 Csapo Tamas Gabor <csapot@xxxxxxxxxxx>:
> Hi,
>
> And is there a reason for choosing 6 kHz in data\scripts\makefilter.pl as a
> maximum voiced frequency, above which the excitation consists of noise? Is
> this setting good for the demo SLT voice only, or is this speaker
> independent?
>
> Best,
> Tamas
>
>
> 2013.02.07. 14:06 keltezéssel, Keiichiro Oura írta:
>
>> Hi,
>>
>> Hmm... You're probably right.
>> I think HTS-2.2 normal demo excitation can be considered as a kind of
>> 'mixed excitation.'
>>
>> Regards,
>> Keiichiro Oura
>>
>>
>> 2013/2/6 Csapo Tamas Gabor <csapot@xxxxxxxxxxx>:
>>>
>>> Dear Keiichiro,
>>>
>>> Thank you very much for your answer.
>>>
>>> In the MELP standard that you cited it is written that "The pulse and
>>> noise
>>> excitation signals are then filtered and summed to form the mixed
>>> excitation." It is clear now that the voiced excitation in HTS-2.2 is a
>>> mix
>>> of impulses and noise. Does this mean that the HTS-2.2 normal demo system
>>> should be denoted / referred as 'mixed excitation' and not as
>>> 'pulse-noise
>>> excitation'?
>>>
>>> Best,
>>> Tamas
>>>
>>>
>>> 2013.02.06. 4:09 keltezéssel, Keiichiro Oura írta:
>>>
>>>> Hi,
>>>>
>>>> To reduce buzzy sound, low/high pass filter are used for excitation in
>>>> HTS-2.2 normal demo scripts.
>>>> You can check the effect by removing following line.
>>>>
>>>> $line .= "$DFS -b $lfil | $VOPR -a $gendir/$base.unv | ";
>>>>
>>>> The filter coefficients are described in
>>>>
>>>> Analog-to-Digital Conversion of Voice by 2,400 bit/second Mixed
>>>> Excitation Linear Prediction (MELP). United States Department of
>>>> Defense. Dec. 1999.
>>>>
>>>> Regards,
>>>> Keiichiro Oura
>>>>
>>>>
>>>> 2013/2/5 Csapo Tamas Gabor <csapot@xxxxxxxxxxx>:
>>>>>
>>>>>
>>>>> Hi All,
>>>>>
>>>>> I have seen that in the HTS-2.2 Normal demo (pulse-noise excitation)
>>>>> some
>>>>> noise is added above 6 kHz in the voiced regions:
>>>>>
>>>>> Training.pl, lines 1958 ... 1966
>>>>>    # synthesize waveform
>>>>>    $lfil = `$PERL $datdir/scripts/makefilter.pl $sr 0`;
>>>>>    $hfil = `$PERL $datdir/scripts/makefilter.pl $sr 1`;
>>>>>
>>>>>    $line = "$SOPR -m 0 $gendir/$base.pit | $EXCITE -p $fs | $DFS -b
>>>>> $hfil
>>>>>>
>>>>>>
>>>>> $gendir/$base.unv";
>>>>>    shell($line);
>>>>>
>>>>>    $line = "$EXCITE -p $fs $gendir/$base.pit | ";
>>>>>    $line .= "$DFS -b $lfil | $VOPR -a $gendir/$base.unv | ";
>>>>> [...]
>>>>>
>>>>> Is there any reason for adding noise to higher frequencies of the
>>>>> pulse-noise excitation? This part of Training.pl wasn't there in the
>>>>> HTS-2.1.1 version, and I haven't found any changelog or paper that
>>>>> would
>>>>> explain why this has been introduced.
>>>>>
>>>>> Thanks and Best regards,
>>>>> Tamas
>>>>>
>>>>
>>>
>>
>