Hi all,


I am wondering how do codecs such as LAME and Speex give better quality results at higher quality settings (complexity settings in the case of Speex)? Correct me if I am wrong but with LAME, I believe it uses a more sophisticated psychoacoustic model to do it right? But how about Speex?. Your helps are greatly appreciated. Also if it is possible, may I have a bit detailed description on it .

Thanks in advance


tech_noobie

what do you mean by high quality settings?, bigger filesize....
Of course, with bigger filesize, more information are stored, better quality.

http://www.hydrogenaudio.org/forums/index....topic=37790&hl=

I am sorry if I am not too clear on what I am saying . There's the quality option in LAME -q so, no I do not mean higher bitrates, the recommended quality setting for LAME is -q 2 and you can get a higher quality output at the same bit-rate by changing this setting to -q 0. Similarly in Speex, there's an option -comp, this allows you to change the computational complexity used during the encoding of your file. At the same bit-rate, higher computational complexity allows you to achieve better quality result, Speex for example allows you to set the -comp settings between 1 to 10. With 4 being the recommended setting and 10 being the one which gives highest quality output at the same bit-rate. The thing that I am asking earlier is, what is being changed at higher computational complexities in both cases. In the case of LAME, for example I believe that they use a different, more sophisticated psychoacoustic model in processing the file - this results in higher computational complexity and thus higher encoding time. I am not sure though whether this is correct or not. Again, in the case of Speex, I have similar question, what enables Speex to give higher quality result at the same bit-rate at the cost of higher computational complexity? For example, does Speex achieve this by applying a more rigorous long term prediction analysis ?(which is what I currently thought at the moment based on my understanding of the code, but ... I am not that good at reading codes, especially C's , thus I am not too sure of it), Is there any other thing that changed at higher -comp settings? I hope I explain my problem better now. Thank you very much for your reply.

In CELP, you try to find the best combination of bits to maximize the decoded quality. Even at low bit-rate like 8 kbps, it still means 160 bits/frame. This is way too much for what can be searched by the encoder, so you have to take shortcuts. The complexity settings actually tell Speex how hard it has to search (or how much shortcuts it can take).

Something similar is true for LAME. It's not so much the psychoacoustic model that gets a bit more complex, but more the "quantization module" that has to meet the demands of the psychoacoustic model with a limited bitrate, which is given more time to look for an optimal solution.