while passing audio through an iir nonlinear phase filter and reconstrucing back actually we alter the phase information eventhough we cant hear bit difference btween the two.why so?

Whether relative phase changes are audible in an audio signal depends on the nature of the phase change and nature of the sample.

I've actually been wondering about this. To what extent is phase relevent to audio?

Canonically, the answer is "not very". In fact, FM stations use phase scramblers for the sole purpose of ripping apart the peaks out of signals, and it's believe to be sort of a free lunch. (cf Orban's article).

There is a great article from the 70s - I cant remember the author. It is the "reference" on audibility of phase distortion.

In essence, they found that low-order phase distortion typically found in loudspeaker crossovers was inaudible in all but a few very special circumstances, such as dry anechoic speech played back over headphones.

-k

There have been a variety of experiments in this regard. Somewhere on this board is a discussion of this subject, where somebody write a matlab script to generate a couple of signals that show that inside of a critical band, in fact phase shift is an issue, IF it's a phase shift that doesn't look like a pure delay.

There have been a variety of reports that "phase shift doesn't matter" that have used many, many degrees of phase shift, but the shape of the phase shift was very close to proportional to frequency, which results in something very much like a pure delay.

On the other hand, 180 degress of phase shift across the sidebands of an AM signal makes it into a narrowband FM signal, and if the sidebands are within a critical bandwidth, the two seem to sound substantilly different.

That's true, it is much easier to hear phase distortions on signals such as sawtooth, while on music signals, same distortions are inaudible.

Of course, phase-linear phase shift does not matter. Or am I misinterpreting you?

The question is how many degrees of phaseshift can be added within a given frequency range before its audible. And what kind of signals stimulates this.

Delaying parts of the audio signal by 500msec would clearly be audible for any kind of music signal with time-information. But the "phase shift problem" of loudspeaker crossoverfilters seems to be exaggerated like many things in hifi.

-k

It is possible to design unusual signals such that about pi/12 over a critical bandwidth (meaning 1 bark bandwidth) or so is an audible phase shift. Such signals are unusual.