[Asterisk-Dev] Re: Handling AST_FORMAT_SLINEAR at 48000Hz

[Steve Kann]

Luigi Rizzo wrote:

>On Thu, Nov 17, 2005 at 10:28:47AM +0000, Tony Mountifield wrote:
>  
>
>>In article <cb2ad8b50511162204i7fcb9a20g78ef72782e80b26f at mail.gmail.com>,
>>Carlos Antunes <cmantunes at gmail.com> wrote:
>>    
>>
>>>On 11/17/05, voipwala hindustani <voipwala at gmail.com> wrote:
>>>      
>>>
>>>>But inside asterisk AST_FORMAT_SLINEAR is assumed to be of 8000Hz sampled
>>>>voice. But in my case it is 48000Hz sampled.
>>>>        
>>>>
>>>You need to do a thing called decimation (or downsampling). The easiest way
>>>is to make the 48kHz signal go through a first order low-pass filter with
>>>cutoff frequency at 4kHz and then collect every 6th sample. You can then
>>>assemble the new stream as signed lineat at 8kHz.
>>>      
>>>
>>The low-pass filter sounds complicated. How would that compare with, say,
>>taking the mean average of every six 48kHz samples to produce one 8kHz
>>sample?
>>    
>>
>
>it's just not the same thing and will give rise to all sort of
>frequency aliasing effects - those that the lowpass filtering is supposed
>to remove.
>quick example: consider a 9khz signal - that is clearly not audible
>on a 4KHz channel (8000 hz sampling rate). Yet if you sample it at
>48khz and take the average of every 6 samples, these averages will not
>be constant (because the 9khz signal does not have a period of 6 samples)
>and so you'll get an output signal that will appear as 1KHz.
>This is the aliasing effect.
>  
>

The filtering or resampling doesn't have to be hard to do. Just go and 
grab the latest source for sox, and use one of the several resampling 
filters that have already been built and tested. You don't need to be a 
DSP guru because you don't need to re-invent the wheel.

-SteveK