[M4IF Technotes] Re: Testing audio Decoder

[Ralph Sperschneider]

Hi Harisha,
Harisha V wrote:
> Hi MPEG experts
> 
>  
> 
> This question related to Testing of MPEG-4 audio decoder. The procedure 
> of Testing of audio decoder for conformance states that compare output 
> of decoder under test with standard decoder for sine sweep and if 
> difference in the RMS lies within range as specified by ISO/IEC ,then it 
> is called as MPEG-4 Audio Decoder with absolute accuracy.

There are a couple of tests specified in the Audio section of the MPEG-4 
Conformance standard. I am aware of at least the following tests:
- RMS/LSB Measurement
- Segmental SNR
- PNS
These tests are utilized depending on the audio object type, e.g. the first and 
the third test are utilized for AAC.
>  What about Testing of other musical test sequences ? Since my decoder 
> can able to decode maximum of 7.1 channels

Only a small number of AAC conformance test sequences is assigned to a dedicated 
test. This is reasonable, since if your decoder passes the test for these 
sequences, it should work properly for any other sequence (with respect to the 
issues tested by this test).
There are a significant number of AAC test sequences with no specific test 
assigned to it, beside others also a number of multi-channel sequences. For 
these sequences, the standard states:
If no test is specified, a check of conformance using appropriate measurements, 
e.g. the LSB criterion (for those sequences that do not utilize PNS) or 
objective perceptual measurement systems, is not mandatory but highly 
recommended. This also applies to bitstreams with non-meaningful window sequences.
So while no test is assigned to those sequences, you should nevertheless test 
them with the same criterion as you do for the sine sweep (RMS/LSB). MPEG does 
not require that a decoder fulfills the conformance criterion for those 
sequences, since some fixed point implementations might have difficulties to 
fulfill this criterion for any input signal.
However, the criterion can be achived, as long as the sequence does not utilize 
PNS. The test output is very valuable, since it allows you to identify serious 
malefunctions (criterion is missed significantly) or rounding problems 
(criterion is missed slightly).
> Is my decoder still be called as decoder with absolute accuracy or 
> should I need to check for all 48+16 channels (48 main and 16 LFE)

You definitely should do this. MPEG simply cannot provide test sequences for any 
and all configurations (although there is a test sequence with 48 main channels 
available).
Best regards,
Ralph
> 
>  
> 
> Can any one answer this question?
> 
>  
> 
> Harish
> 
>  
> 

-- 
Dipl.-Ing. Ralph Sperschneider  | Phone: +49 9131 776 344
FhG IIS-A                       | Fax:   +49 9131 776 398
Am Wolfsmantel 33               | mailto:sps   iis.fhg.de
D 91058 Erlangen                | http://www.iis.fhg.de/amm/