Tuesday, January 29, 2008

What's The Theory?

Very quickly: If Romney has a handler talking to him through a wireless mic, that signal could have been picked up by one of the studio antennas. It would appear to be coming from whatever microphone was assigned to that antenna. (Some people are saying it would be impossible to pick up the ear piece into a mic -- essentially true, but the wireless signal could be picked up.)

Monday, January 28, 2008

Agressive Processing

Enough of the mild stuff -- I want to know what's in the "gun whisper" clip, so I've processed the sucker pretty heavily. I also realized that the utility I used to rip the original track was upsampling the signal -- poorly! So I made it stop, and that's given me a bit of an improvement.

Here's my running theory about the gun clip. You can tell Romney is choosing his words carefully, which I suppose one has to do when explaining why you were for new gun legislation before you were against it. He tells us "I said I'd -- I would support" a particular new gun law at the federal level. The you have this:

Romney: "I do not believe we need new legislation"

Whisper: "Support"

Romney: "I do not support any new legislation..."

Is there a difference between believing and supporting ... maybe so. Maybe "support" only comes into play when there is a particular bill that's already been proposed. At any rate, listen to the (very heavily processed) clip and see whether you believe there is a whisper saying "Support."

The lower frequencies, where most of Romney's voice is, were filtered out, then the entire signal was shifted down (not a change in sampling rate or speed, though). The results are A) it plays funny to the ear, because the harmonics are no longer harmonic. B) The supposed whisper is louder relative to Romney's voice. C) The whisper is lower in frequency than in the original recording.

And, just for fun, I've done similar processing on the "He raised taxes" whisper:


Sunday, January 27, 2008

Updates On Both Whispers

I've posted the 'click' processing on YouTube ... it's nothing revolutionary, since the click can be heard on the unedited video, but it can be heard better here, and the whisper is easier to make out as well. Video will play the unprocessed audio first, then the processed audio. There's another video out where the whisper has been enhanced, but that one adds some other artifacts that can confuse the subject of the click.

I've also processed the whisper heard during the gun question. I've processed this many different ways, and I'm not sure that it's a whisper. It's the right frequency, and looks about right (very similar to the audible "taxes" whisper), but when it's isolated, it doesn't sound much like a whisper. Here's the chart:

Region A is the "ation" part of "legislation." B is the word "I" from "I do not support..." Region C is the part which, when played, sounds like an whispered "s" possibly from someone whispering "support." I compared that whispered "s" with the one from "taxes." The frequency doesn't match all that well -- but it's possible -- of course, it could be a different speaker as well.

Here's the whisper sound from the gun question, processed and unprocessed. I've isolated it, and I think anyone would agree that it's unconvincing:

Both of the clips I have were compressed in the original format. That means information was lost that cannot be reconstructed. I'd like very much to get uncompressed and multi-channel signals. There are a few more tricks I can try, but I doubt there's much more to get from these files.

Call On MSNBC To Release Raw Audio

The audio that I have available is compressed and single-channel. There would be a lot of evidence available in the uncompressed multi-channel audio. Especially given that NBC reported that the whisper was only in one channel. That's an important clue, and may make it impossible to blame this on an open mike in the audience. The 'audience' theory would seem ludicrous anyway -- the moderators did their best to suppress audience noise, and indeed very little was heard, yet this one whisper is crystal clear -- no coughs, shuffles, chatter -- just this one quite meaningful, ready, and well-informed whisper.

These debates are a significant factor in determining arguably the most powerful person in the world. It is a sacred public trust. If there's a suspicion that the debates are not quite what they appear to be, MSNBC ought to do what it can to put this issue to rest. MSNBC should release the raw audio. Let them know how you feel.

Saturday, January 26, 2008

Two clicks

UPDATE!! see below

Also see later post here

OK, so you've heard the Romney whisper. Now take a look at the two 'clicks' that come on either side of the whisper.

The following graph shows the FFT (log) of the segment. A Hanning window was used, for those who are interested, 1000-pt, 50% overlap window-to-window. Nothing crazy.

The area denoted "A" is the whisper. It's pretty typical of a whisper -- high frequencies, but no definite frequency pattern, no clear pitch. Whispers are essentially noise (think 'sssss') passed through filters to shape them.

The area denoted "B" is a 'click' sound. It's quite audible. Visually, you can see that there is no pitch, and a fairly even frequency distribution. It starts and stops suddenly. It's a click. Note that it occurs after the whisper. A microphone is clicking off.

Is there an 'on' click. Yes! Area "C" is the tail end of Tim Russert's question. It's typical resonant speech. It's the middle-to-end of the word "three" (1983). "Three" has resonant sounds; there are no 'clicking' or what a speech processor would call plosive sounds in the word. You can see a very clearly defined pitch, along with harmonics.

But there's an additional sound in there, area "D." It has evenly distributed frequencies, like the "B" sound. The human voice is not capable of making a clicking sound and the "ee" sound at the same time. This sound comes from a second source. I can't make it out on the audio, because the ear is much better at picking up resonant sounds -- Russert's voice, in this case. But it's there. It's also a click. It's the microphone clicking on.

UPDATE: Romney is known to have worn an ear-piece while speaking at the American Dream Summit in Michigan. One of his handlers was at the other end:

UPDATE 2: Well, we need to all try to be fair and honest, even in the pajama media. When I went to further process the audio, I realized that I had not applied the Hanning window, as was stated in the text above. The Hanning window is applied to get rid of certain FFT artifacts, and the appearance of the first 'click' is a possible artifact. (The FFT treats a signal as though it were an endless loop, and looping can cause a discontinuity, much like a click, that isn't really there.) So, I fixed that, and the first 'click' disappears:

So there's a click after the whisper, but if there's one before the whisper, it's covered by Russert's voice ... at least for now.