Dr Klein was able to send us a baffling case of paranormal voice intrusion that happened during a Skype conversation running between Dr. Emanuel Toriello (ISARTOP chairman) and himself a year ago or so. It is worth mentioning that Dr Toriello has experienced this kind of anomolous interference several times. Listen to sound file (1 min). Dr Klein submitted the file to EVP expert Daniel Gulla who concluded that the voice was not of human origin...read his report.


ANALYSES OF "PARA-INTRUSION" SAMPLE
Daniele Gullà

SUBJECT:
Analyses performed on a sample recorded by Dr. Adrian Klein from Israel concerning a noise interference, similar to a verbal elocution, introduced during a phone conversation through Skype communication system.

The electroacoustical analyses have been performed using several software. Such analyses revealed the interfering signal much more similar to a modulated noise than to a human vocal articulation. In particular the sound seems to be modulated in high wide frequency ranges with a constant frequency change assimilable to a sound produced by a synthesizer.

In the following some spectral analyses and main characteristics are reported.

Fig. 1

The violet portions of oscillogram in Fig.1 are related to the mysterious interference.
On top-right a sound coloured intensity (dB) spectrogram, with Waterfall projection, is represented. On bottom-right is shown the Speech Intelligibility analysis, based on seven frequency bands.

It is pointed out as the signal is limited in a constant band with a maximum of energy ranging from 1 to 4 KHz. With the exception of some isolated peaks, it is not revealed any signal at low frequency.
Presuming the sample to be a vocal articulation the signal intelligibility is almost equal to zero (measured 0,01 Bad).

Fig. 2

By expanding the spectrogram scale to 20 KHz (the previous one was set 0÷8 KHz), some bands with minor energy are detected, 8÷12 KHz and 14÷16 KHz respectively.

The energy trend is undulatory, as shown in the spectrogram, and similar to the one produced by a sampler and an electronic synthesizer.

-- Voice report for Sound Paraintrusion --
Date: Mon Apr 3 21:45:54 2006

Time range of SELECTION
From 53.061003 to 69.128389 seconds (duration: 16.067386 seconds)
Pitch:
Median pitch: 274.774 Hz
Mean pitch: 305.748 Hz
Standard deviation: 113.124 Hz
Minimum pitch: 137.250 Hz
Maximum pitch: 499.675 Hz
Pulses:
Number of pulses: 482
Number of periods: 432
Mean period: 3.362400E-3 seconds
Standard deviation of period: 1.358713E-3 seconds
Voicing:
Fraction of locally unvoiced frames: 74.860% (1203 / 1607)
Number of voice breaks: 27
Degree of voice breaks: 84.930% (13.645969 seconds / 16.067386 seconds)
Jitter:
Jitter (local): 6.854%
Jitter (local, absolute): 230.451E-6 seconds
Jitter (rap): 3.613%
Jitter (ppq5): 4.314%
Jitter (ddp): 10.838%
Shimmer:
Shimmer (local): 15.891%
Shimmer (local, dB): 1.380 dB
Shimmer (apq3): 8.867%
Shimmer (apq5): 11.490%
Shimmer (apq11): 14.880%
Shimmer (dda): 26.601%
Harmonicity of the voiced parts only:
Mean autocorrelation: 0.598294
Mean noise-to-harmonics ratio: 0.713393
Mean harmonics-to-noise ratio: 1.809 dB

The analyses reveal the impossibility to classify the sample as a natural voice since the 74% of sound articulation is not recognizable as a voice. In addition the Signal/Noise ratio (1.8 dB) is not acceptable for an accurate analyses of the electroacoustical parameters like formants and for calculating the relevant the Linear Prediction Coefficients (LPC) and Cepstrum.

Fig. 3

Fig. 3 shows the F0 Statistic report extrapolated from a perceptive tonic vowel \A\. The relevant values are rather stationary and the presence of F0 is poor and fragmented; as consequence such material is insufficient for a linguistic analysis.

By using "Dr. Speech" software (recently acquired by Il Laboratorio) only 0.72% of F0 can be considered within the limits of parameters pertinent to a male speaker.
In the oscillogram shown in the bottom of Fig. 3 (represented in red), it is evident the lack of F0. When F0 is detected the oscillogram results coloured in green.

In Fig. 4 the pitch, marked by F0 in very short while only, is analysed. It is characterized by some heavy tone variations ranging between 1.5 and 7.6 ms.

Fig. 4

The pitch analysis evidences, in some frames only, high temporal changes. The pitch, as reciprocal of fundamental frequency F0 produced by the vocal cord vibrations, is represented as average in milliseconds on left vertical scale.

Fig. 5

By using Dr. Speech software the analyses on the vowel perturbation quotient resulted in the impossibility to be accomplished due to the indefinableness and fragmentation of F0. In addition such impossibility is due to the mixing of tonal component of vowels into the noise in consequence of the particular communication channel or the intrinsic nature of signal (see warning advice on screen in Fig. 5).

By accomplishing the formant analyses, in Fig. 6 is shown the IPA table with vowel dispersion area. All vowels are located in the front labio-postalveolar and labio-palatal areas.

Next Fig. 7 shows the analysis performed, using wavelet modality, on a sample fragment where a noise with a "small waves" pattern. It is highly similar to a signal processed by a vocal synthesizer.

In the following a statistical calculation of formants F1/F3 changes in the time (variable 1) is reported. Such formants result to be uncertain and stationary.

Fig. 6

Fig. 7

CONCLUSIONS

The instrumental analyses performed using data processing systems evidenced the only presence of physical characteristics uncommon in a human voice like vocal cords frequency F0 and trajectories of formants Fn.

The presence of heavy noise in those bands (Fn) causes uncertainties in the values distribution then difficulties arise to choose suitable specific algorithms, like Cepstrum on formants, required to reach a clear spectral view necessary to extract the characteristic frequency range of each selected phoneme.

The detected anomalies, like F0 values and formant statistics, may be considered as intrinsic peculiarities of the analysed signal or, taking into account the heavy spectral deterioration, similar to the one used in many processors for acoustical effects (synthesizers), such signals may be also considered as an acoustic sample artificially manipulated.

By considering the impossibility to perform an accurate parametric analysis, the doubt if the signal analysed may be considered as a processed signal by a synthesizer or an EVP still remains.

I don't have patterns with expansion of variables for analysing the particular characteristics of synthesised signals, anyway, for what I know, it is my opinion that it is possible to create, by using a synthesizer, vocal spectra more similar to the human voice in spite of what I found in the above analysed sample.

The Technical Consultant
Daniele Gullà

<< Return to Index