Hi,
I downloaded AMR-NB/WB code from ITU-T website for development.
I compiled AMR-WB and tested with a 4 second speech stream @16KHz and 16bits/sample.
Using mode0 i.e. 6.6kbps mode it gives me encoded file as ~53KB size
BUT
I expect an encoded size of (6.6kbps*4*1000/8) bytes i.e. 3300 Bytes or ~3.2 KB
Again, input bitrate is 256kbps to the encoder and we are supposed to get 6.6kbps i.e. a compression gain ratio of ~38-39 but if we consider 53KB encoded stream as correct then the corresponding bitrate is ~108kbps which is a compression gain of 256/108 = 2.3 which seems to be very less for a speech encoder.
this is happening with all the mdoes...
I need to be clear on this to proceed further to development of the codecs.
AMR-NB also has same problem...
Please let me know the testing procedure and if I am missing out anything before I start.
Thank you.
The ITU-T code of G.722.2/AMR-WB uses the G.192 way of representing bits (check G.191 for more details). Actually, if I am not mistaken, most if not all codecs in ITU-T use that representation. It is just the way bitstreams are stored and does not reflect in anyway the compression that is done.
Basically, the bits are "soft" bits. Each frame of data is composed of a sync_word (16bits) the length of compressed data frame (16 bits) and the compressed data where each bit is represented by a 16 bit word.
In your example:
(16 bits for each bit) + (ovehead 32 bits per frame)
6.6kbps*4*1000/8 * 16 + (4*50*32)/8 = 53600
Hi,
I downloaded AMR-NB/WB code from ITU-T website for development.
I compiled AMR-WB and tested with a 4 second speech stream @16KHz and 16bits/sample.
Using mode0 i.e. 6.6kbps mode it gives me encoded file as ~53KB size
BUT
I expect an encoded size of (6.6kbps*4*1000/8) bytes i.e. 3300 Bytes or ~3.2 KB
Again, input bitrate is 256kbps to the encoder and we are supposed to get 6.6kbps i.e. a compression gain ratio of ~38-39 but if we consider 53KB encoded stream as correct then the corresponding bitrate is ~108kbps which is a compression gain of 256/108 = 2.3 which seems to be very less for a speech encoder.
this is happening with all the mdoes...
I need to be clear on this to proceed further to development of the codecs.
AMR-NB also has same problem...
Please let me know the testing procedure and if I am missing out anything before I start.
Thank you.
try Voice Age:
http://www.voiceage.com/freecodecs.php (http://www.voiceage.com/freecodecs.php)