Skeptics, who are looking for conventional "spirit voices" in Séances, usually assume that the medium itself pretends such voices (even if media such as Warren Caylor lock their mouths tightly with tape or gags). Although the Spirit voices sound differently from the voice of the medium, it is difficult to decide whether a Spirit voice has been deceived on the basis of such subjective judgments. Here, techniques of speech signal processing can help to provide objective evidence and, at the same time, use speech features that are hardly (consciously) audible as they are not relevant to our understanding of the language.
Here is some information and preliminary results of the examinations using forensic voice comparisons based on audio recordings in Séances with Warren Caylor. Thanks to Warren and his Spirit Team for the support of this work!
Note: These are ongoing investigations / evaluations, so the results and descriptions on this website are very preliminary. In the last section, I explain further, still pending work steps to deliver more robust results.
Can the medium simulate the different Spirit voices?
A first, obvious question of how to deal with Séance skeptics is: How likely it is that the medium is capable of simulating the various voices heard during the séance, taking into account that the medium has its own Voice would probably deliberate.
Here are some representations of different Spirit voices from Warrens Séances:
For comparison, a professional "musician"
In order to assess the diversity of Spirit voices, a central question is the extent to which a speaker (the medium) is able to make his or her voice sound differently. For comparison, I have looked at the recordings of a professional artist, who is consciously working with his language. In his ("Kangaroo Chronicles"), Marc-Uwe Kling performs extremely lively dialogues between himself and a kangaroo Himself speaks clearly, but his voice clearly distorts his voice.
For more reliable documents, of course, further measurements and objective distance metrics are desirable. But these first results also suggest that the voices to be heard in Warrens Séancen are significantly different than a stage profi can achieve with his own voice (without having closed the mouth with tape, as was the case with Warren Case).
Spirit voices of famous personalities
Other interesting questions arise if the Spirit voices are to belong to well-known persons, of whom audio recordings are available from their lifetime. Since Warren's Séances regularly shows a Spirit as Winston Churchill, there are additional interesting research possibilities.
In such cases, four or more different voices may be of interest to the analysis:
The original voice of the medium
The Spirit voice during the séance
Original recordings of the real, historical person during their lifetime - possibly at different ages, as the voice may change over the years
Other people, especially actors who try to imitate the voice of the original person
It is clear, in this case, we are clearly going further from the suspect / perpetrator model of the typical forensic vote comparison so that the classical probability ratio used there would also be extended.
Here are some preliminary evaluations of the voices. The first shows the Spirit voice of "Winston" in Warrens Séances, the second the original voice of Winston Churchill from the (arbitrarily chosen) Youtube video "We shall fight on the beaches" and the last the actor Albert Finney as he Winston Churchill in 'The Gathering Storm' plays:
In these comparisons, the problem arises that the stimuli are taken from different sources and, in particular, original recordings of Churchill are of a significantly poorer quality. It remains to be examined to what extent this is reflected in the evaluation of formants. It also raises the question of how Churchill's voice changed in the course of his life. Maybe the Spirit voice corresponds to a younger or (as Lucius Werthmüller suspected) older Churchill. At present, I have only used comparatively short stimuli, so that even more comprehensive evaluations are required. From the first impression, perhaps,
These are preliminary results! Comments and next steps ...
Again, it should be recalled that these are provisional results. Further investigations are necessary in order to collect hopefully more robust evidence. Some possible next steps:
Calculating an objective metric to measure the distances / differences between the formant graphs, typically to calculate a probability ratio, as is currently the best practice in the forensic vocal comparison.
Collection of more extensive data.
Examination into the extent to which the different audio quality can distort the results in the Churchill evaluations.
Marc-Uwe Kling and his kangaroo are just an example of how shifting the voice can affect. Further examples would be useful to underline this approach with more data.
With regard to the formants, their trajectors (ie the time-dependent courses) could provide further material for the speaker identification. This, however, requires the (tedious) work of identifying and setting up individual sounds in the audio recordings.
The formant analysis was done with the software Praat, which is very popular in this area, with my own software taking some steps for postprocessing, eg to generate the correlation diagrams and to filter out isolated measurement errors (Praat designates F2 as F1 if F1 was not recognized correctly). For more precise results, the automatically calculated formants should be manually checked by visual inspection of the spectrum and corrected if necessary (tedious work ...).