Downey Synthesized Speech EVP

recorded by Margaret Downey
©Margaret Downey – All Rights Reserved

Speech Synthesis can be tricky to work with, so please be sure to use a witness panel.

As the technology of EVP recording evolves, new techniques have brought new forms of the phenomenon. The most common and best understood is the technique using a recorder, microphone and supplied noise for voice formation. It is theorized that actual voice formation is accomplished by amplifying a small telekinetic signal in the active region of transistors via stochastic resonance acting on ambient noise energy. This is a hypothesis, but it answers most of the evidence. We refer to this form of EVP as “Transform EVP,” and most of the examples on this website are of this kind.

A currently popular technique for EVP collection is referred to as “radio-sweep.” This may be automated or manual, and depends on the live voice of broadcast radio as a sound source. Small fragments of sound are formed into a new sound stream by rapidly sweeping the radio dial. This form of EVP may be referred to as “opportunistic EVP” because it depends on the availability of sounds necessary to form the message just when they are needed at the moment each station is selected.

A second approach to opportunistic EVP is EVPmaker. This computer program loads an input sound file into a buffer, and then using a random process to determine which part of the buffer to use as the next fragment of the output sound stream, forms a new sound stream. Several “random” sources are available and the fragment length is determined by the operator. Assuming prerecorded voice is used, the output is a staccato sound that contains randomly arranged voice fragments. Meaningful messages are sometimes found in the output, and it is thought that the communicating entity is influencing the random process to access the buffer to select the needed sounds.

A great deal is known about transform EVP and false positive results are usually easily detected. Radio-sweep and EVPmaker pose the problem that normally occurring words and phrases in the live voice are easily mistaken as phenomenal message, and use of live voice makes these technology unlikely candidates for research. See Locating EVP Formation and Detecting False Positives for a discussion of this problem.

A new technique for voice formation involves speech synthesis, rather than live voice. For instance, Stefan Bion introduced a set of SpeekJet allophones on his website. These are fragments of sound used for the synthesis of speech for robotics. A few words may naturally occur when the allophones are used with EVPmaker, but in general, if a word is formed and it seems to be meaningful, then it is unlikely to be a natural product of the technology.

Real-Time Conversation Using EVPmaker and Allophones

Margaret Downey has recorded convincing examples of EVP using just about every technique she has tried. In the example provided here, she used the allophones with EVPmaker in an attempt to conduct a real-time conversation.  The output of EVPmaker is a steady staccato string of pieces of voice. Margaret spoke over this, so that you can hear her voice and the EVPmaker output in the background. She then lowered her voice in the recording and removed the non-relevant EVPmaker output for clarity. Here is the script:

Margaret: Can you say A.B.C.?

    EVP: A.B.C.D.E. (this is buried in the main sound)

Margaret: How about Tom Butler?

   EVP: Tom Butler.

Margaret: Do you know the name of either of Tom Butler’s cats?
   EVP: (Sh)asta’s their cat. (the “Sh” sound is weak or missing)

“Shasta” is the name of one of the Butler’s cats.

Margaret: Do you have any pets?
   EVP: I have a pet.

Margaret: I heard you say I have a…?
   EVP: Pet.

Margaret: You have a what?
   EVP: Pet.

Margaret: What kind of pet do you have?
   EVP: An animal. (Ask a silly question, get a silly answer!)

Margaret: I understand you have a pet animal, but what is the species?
   EVP: A purple deer.

Margaret: OK, I could have sworn you said you have a purple deer.
   EVP: I told you, purple deer.

Margaret: OK, I’m gonna close for the evening. Can you tell me goodbye?
   EVP: I love you.

Margaret: I heard you say I love you. Love you too.
   EVP: Thank you.

Conversation With Arthur

It is important to note that it is common in in transform EVP to recognize the voice of the person thought to be speaking, sex and approximate age of the speaker. This added information helps people recognize loved ones, and a message from a loved one is often very therapeutic. The output sound stream of EVPmaker can be transformed into voice as in transform EVP, but it is more common for EVP made with current applications of speech synthesis to be a more robotic voice with little personality.

Another point that needs to be made is that, even though the voice is an emulation of live speech made by transforming noise into voice, transform EVP often includes the nuances of speech, such as plural and accents. Opportunistic EVP is more often an approximation based on available sounds that are close to what is needed. Thus, “I love you” might actually sound like “eh ove u.” In speech synthesis, the word can be more closely matched, but we do see novel spelling when the words are translated to text.

Real-Time Communication Using the Paranormal Puck and Allophones

A recent innovation in technology for EVP is the introduction of the Paranormal Puck by Digital Dowsing: Owner and developer is Bill Chappell. This device connects to a Windows computer via a USB port. It includes temperature and electromagnetic sensors and a SpeekJet microchip. In one mode, the output is an environmentally stimulated string of allophones.

Margaret Downey conducted sessions with the Puck by asking a question, and then turning on the puck for thirty seconds. She then turned off the puck and asked another question for repeated sessions. The result is the appearance of conversation, but be aware that this is a very new technology and it is difficult to know what is actually occurring. The impression is that questions are ask and correctly answered.

In this example, Margaret had the device turned on but had not decided whether or not she was going to conduct an experiment so she had said nothing.

Puck: Talk to him.

Margaret: In this I asked, “Please say my grandson or granddaughter’s name.” (Dylan Scott or Hazel Ann – both on the other side)

Puck: Scott

Margaret: This wasn’t the answer I expected, but I like it, “Who are the babies with?” At 16 seconds, I heard, “Margaret” followed by “Dylan” at 23.5 seconds phoneme: ovrat (for Margaret) oops, forgot to note what it was for Dylan. Timing altered.

Puck: Margaret, Dylan

Margaret: The previous three questions where answered the first time I asked. This was the third time in a row I had asked this question (and would have been my last attempt before moving onto the next question.)

It is important to remember that the technology being used in these examples should not normally produce recognizable speech. Yes, there is the probability of an occasional string of sound bits producing a recognizable word. A good comparison is that a thousand typing monkeys will eventually produce something meaningful; however, what is the likelihood of the monkeys producing something meaningful that is also a correct answer to a question just asked?

Leave a Reply