The second screen shot shows two more tests. The first is human speech (letters A,B,C,D). The second is knocking of the knuckles on a table top. Who wudda thunk, huh?
Deciding what part of a waveform can be considered redundant and therefore safely removed, without perceptible degradation of voice quality, was once part of our very frustrating research in voice compression techniques. Various models had ben tried - some mathematically sophisticated, some simplistic.
Of the latter category was this brute force approach: record few dozens of seconds of a person's speech, break it into thousands of vectors of the same size, then choose some number of them, 1000 say, that would best represent the entire speech sequence. This was a so-called training session where some pattern matching techniques were used to make the pattern table, or the dictionary - if you wish.
The dictionary was then to be used in the next step - in the real time voice compression session, where the same person would speak, vectors of his/her voice extracted, matched to pre-stored dictionary vectors, and the resulting keys (actually numbers) sent over to a receiver.
On the receiver side, the same dictionary was used to decode the original speech by converting keys to values (vectors) and piecing all those vectors together. Again, this was done in real time.
Quality of the resulting synthesized voice depended on many factors - dictionary size and vector size being the major ones. Obviously, the bigger the dictionary and the smaller the vectors the better the approximation of the original - but, on the other hand, the lower voice compression ratio.
We tried using a dictionary of a person A during speech compression session of a person B. To our surprise this often worked reasonably well, even if A was a soprano woman and B a baritone man.Since I quit that research long time ago, I am still clueless to this day what voice features are essential and which part of voice can be considered redundant. Some people obviously do know that, since voice compression techniques are everywhere nowadays: telephony, satellite communication, voice over internet, etc.