Hi,
There is an algorithmic delay when u decode with the speex.....
basically, if u encode a 'abc.wav' file to 'abc.spx' and then decode it back to some 'xyz.wav', u get a delay in at the starting in the 'xyz.wav' compared to 'abc.wav'
I did some experiments with a 16KHz speech file....
by default speex encode it at a default bit-rate(i.e. 27800 bps in my case, I saw it thru Verbose mode) I dont get a delay, i.e. the files 'abc.wav' and 'xyz.wav' are aligned sample by sample and I get a fairly small residual...
But when I force it to encode at a bit rate of 13000 bps and 6500 bps, I get a delay at the starting....
In the case of 6500 bps I get a delay of around 1ms.
Could you please explain it to me why there is no delay in the default mode and there is a delay at lower bit-rates.... (I didnt check for a bit-rate higher than 27800 bps)
Are you using the latest code? Only newer versions have sample-exact lengths or gapless playback.
If you are using the latest code, maybe there's a bug with some of the modes, perhaps?
What did you use to encode?
Are you using the latest code? Only newer versions have sample-exact lengths or gapless playback.
If you are using the latest code, maybe there's a bug with some of the modes, perhaps?
What did you use to encode?
I am using
speexdec (Speex decoder) version 1.0.5 which I downloaded from speex website two months earlier.... I think this is the latest version.. or any other latest thing is there ??
One more thing I want to ask is this that when I dont use the default mode, even if I shift and align the reconstructed wavform.... the residual variance is not considerably low unlike in the case of the default mode where its 100 times smaller than the variance of the original wavform....
No newer binaries are at speex.org, but you can get the source for 1.1.10 from Xiph SVN. http://svn.xiph.org/ (http://svn.xiph.org/)
I *think* that 1.0.5 binary has gapless decoding, but I'm not sure.
But what did you use to get the .spx files? 1.0.5 lists gapless playback as one of its features, but it doesn't mention gapless encoding. 1.1.6 lists "Gapless playback for speexenc/speexdec" so I guess it has gapless encoding.