Hearing ghosts? EVP and the DR60

2024 note: I left the May 2022 write-up below unchanged when I added the Halloween 2022 update. They both should be read to understand my final conclusions on this issue.

May 2022: a surprising experience

In May 2022, my wife Sally and I were able to take part in some paranormal investigations at a location not too far from our Washington DC home. Some of these involved “Electronic Voice Phenomena,” or “EVP,” which is the recording of unexplained voices on electronic recording equipment.

I was very surprised by what I witnessed. In a sequence of tests with a handheld voice recorder, in which each recording was done for a minute or so and then immediately played back, in essentially every instance there were extra loud sounds on the recording that were not audible when the recording was being made. These were not subtle sounds: they were loud, distorted, almost “barking” sounds—louder than any of our voices, and at times during which there was essentially absolute silence in the room. Since each recording was relatively short (a minute or so), there was no chance that we had simply forgotten about passing noises. These were loud sounds that definitely were not audible when we were recording.

I knew little of EVP before these investigations. From the parts of the few “ghost hunting” shows that I had seen in passing on cable TV, a few had included what I now knew was called “EVP,” but I had always assumed that the sounds had actually happened in the room at the time—just that no one had noticed it, or it was too subtle or quiet to notice at the time. Now I learned that my impression was wrong: these are sounds that were not at all audible at the time, but somehow ended up on the recordings.

I was intrigued. Another recording from the same voice recorder, made before we were present, was played back to us. This recording was of more normal-sounding voices. In this case I noticed that the audio was quite poor: it had what seemed like quantization errors (clicking and buzzing) and clipping (distortion when the sound level was too high)—but, most interestingly of all, it had aliasing, which is when high frequencies are “folded over” to lower frequencies, resulting in a distorted sort of output that is unmistakable once you know what it sounds like.

I noticed that the voice recorder seemed quite old, and wondered if this might have something to do with whatever it seemed to have been capturing. Later on I asked what exact model it was. It turns out it was a Panasonic IC Recorder RR-DR60, manufactured in the 1990s. It was apparently so legendary in paranormal circles that it was referred to simply as “the DR60,” and trying to buy one secondhand was close to impossible.

The DR60

The Panasonic IC Recorder RR-DR60

Hypotheses

By the next morning, I had decided that there were likely four broad possible explanations for what we had witnessed:

  1. The DR60 software had bugs in it, resulting in it making up sounds that never entered its microphone.
  2. The DR60 was not filtering out high-frequency or ultrasonic sounds, and these were being aliased down to normal voice frequencies.
  3. The DR60’s electronic circuits were picking up electromagnetic interference, effectively converting the induced currents into sound signals.
  4. Paranormal phenomena were able to rewrite the bits and bytes in the digital storage of the DR60, implanting sound messages there directly.

Note that Options (2) and (3) do not address the question of whether the ultrasonic or electromagnetic signals potentially causing the anomalous recordings were of normal or paranormal origin; both are theoretically possible. Option (1), on the other hand, would rule out any such EVP from being evidence of paranormal phenomena, whereas Option (4) explicitly assumes it.

Assessing the hypotheses

Considering the four hypothesis above, I made some immediate subjective assessments.

Option (4) seemed to be not worth further exploration, for two reasons. The first was practical: if the signals were not obtained by any physical process, then there would be effectively no way to analyze or study them. The second was more philosophical: if paranormal phenomena were able to write bits and bytes directly, why would there be any need for an electronic voice recording device at all? They could just as easily write these bytes directly into the memory of a computer or music playing device. I therefore laid Option (4) aside, as something that I could not sensibly investigate.

Option (3) seemed to me to be the most likely obvious explanation. Any recording device is sensitive to the small currents induced by its microphone(s), and any electromagnetic waves strong enough to penetrate the electrical shielding may induce currents that could feasibly sound like those barking sounds.

I did some googling, however, and it seemed that this most obvious explanation had been proposed and addressed numerous times. Voice recorders—including the DR60—had apparently been put into Faraday cages, and still produced the same EVP recordings. Although I was skeptical of any experiments that did not seem to be fully documented or carried out with scientific rigor, I thought that it was possible that Option (3) might not be the actual explanation. I put it aside, with the intention of returning to it if the other options were ruled out.

Option (1) didn’t seem to be the most likely explanation. It was difficult to think of a software bug that would simply insert barking sounds that were not there, rather than fully distorting or corrupting all recordings made by the device.

Although that wasn’t a convincing argument against Option (1), I also put it aside—to be resurrected later if no other options were to pan out—for the simple reason that, absent a copy of the source code for the software on the DR60, it would be effectively impossible to determine if it contained such a bug.

Option (2) seemed to be the most likely explanation to me, since I knew from direct listening that the DR60 suffered from audible aliasing of high frequencies. I decided that I would investigate Option (2), as best possible.

Investigating Option (2)

I figured that investigating Option (2) would require at least four things:

  1. A DR60.
  2. A device capable of producing high-frequency and ultrasonic audio.
  3. A recording system capable of faithfully recording that high-frequency and ultrasonic audio, so that it could be compared with the recordings of the DR60.
  4. Audio analysis software capable of handling high-frequency and ultrasonic audio.

Item (4) turned out to be easy: Audacity is a free and open-source application more than capable for the job, handling audio with sampling frequencies up to 384 kHz.

Item (3) was not too difficult to find: the Pettersson u384 Microphone is a relatively cheap USB ultrasound microphone that samples at 384 kHz, with a frequency response down to around 1 kHz, sold for measuring the echolocation sounds of bats. It was also easily ordered in the US from batmanagement.com.

Item (2) was similarly straightforward: the Peersonic Ultrasound mic tester is relatively cheap and generates test ultrasound sequences between 14 kHz and 125 kHz, and it was also easily ordered from batmanagement.com.

Ironically, the most difficult item to obtain is Item 1: a DR60. They were apparently only manufactured in the 1990s and early 2000s, so only second-hand devices are available. But their popularity for ghost-hunting has caused them to be priced between $1,000 and $5,000!

Although I was eager to test out my hypotheses, I couldn’t justify paying this sort of price for something that probably sold for $50 when it was released.

Although I put out feelers to those in the Washington DC region for anyone who might allow me to test their DR60, I decided that in the mean time I would look for alternative ways to test Option (2).

Alternative EVP recorders

I knew that the DR60 wasn’t the only digital voice recorder claimed to be useful for EVP—only that it seemed to be the best. A number of other such recorders were listed by people who had performed paranormal investigations, and apparently obtained good EVP recordings from them. My logic was: if this is not just an artifact of the DR60 alone, then the same sort of behavior should be shared by at least some of these other recorders. It would make sense that aliasing would occur with earlier voice recorders, since I had heard it first-hand on the DR60.

Surveying the published opinions of those investigators, and looking for devices available, I ended up ordering four recorders, of different vintages:

Panasonic IC Recorder RR-QR240

This device seems to be one of the next generation of digital voice recorders released by Panasonic after the DR60. It has 240 minutes of recording time (compared to 60 of the DR60), and looks somewhat similar, but has significant differences. Like the DR60, it does not have any digital output, just a 3.5 mm headphone output jack (the DR60 has a 2.5 mm output jack).

The QR240

The Panasonic IC Recorder RR-QR240

Olympus Digital Voice Recorder WS-801

This device from Olympus is another generation on again from the Panasonic devices. Unlike that earlier generation, it has a USB-A plug (which slides out), and uses both the MP3 and WMA audio compression standards.

The WS-801

The Olympus Digital Voice Recorder WS-801

TASCAM DR-05

This relatively modern recorder has since been superseded by an improved model (the DR-05X), but is listed as a good EVP recorder.

The TASCAM DR-05

The TASCAM DR-05

Zoom H4n Pro Audio Recorder

Finally, this is a current high-end audio recorder.

The Zoom H4n Pro

The Zoom H4n Pro Audio Recorder

Investigating Option (2) on the alternative recorders

It didn’t take me long to discover that, not only was there was no appreciable aliasing for the Zoom H4n Pro or TASCAM DR-05—as expected, for such modern devices—but there was no appreciable aliasing for the Olympus WS-801 either. This surprised me: Option (2) relied on the guess that older digital voice recorders likely did not filter out higher frequencies rigorously, which could be aliased down to normal voice frequencies. But this did not seem to be the case, at least for the WS-801.

I therefore tested the Panasonic QR240 with bated breath, as this was the closest device that I had to a DR60. In this case, I did detect some aliasing of high audio frequencies, but the aliasing did not fold down appreciably into normal voice range (i.e. that could represent the “barking”), I didn’t detect any appreciable aliasing down of ultrasonic frequencies, and the intensity of the aliased high-frequency audio signal was far too low to get anywhere near explaining the loud sounds that we had heard on the DR60.

These results were disappointing in the context of Option (2), but I suppose were a testament to the fact that the designers of these digital voice recorders did a better job of filtering out high frequencies than I had guessed. (My heartfelt apologies to these engineers for casting aspersions.)

I was now in a quandary. I knew that the DR60 had aliasing artifacts, but would they be strong enough to create the “barking” sounds at relatively lower frequencies? And could the DR60 really be alone in reacting in this way?

I tried finding as many technical details about the QR240 and DR60 as I could, and chanced upon a document that may explain everything.

Discovery of a new document

Looking through the manual that I had found online for the QR240, I noticed that in the Specifications section it specified: “Frequency response: 450 Hz – 5.0 kHz.” This surprised me, because I remembered reading that the DR60’s manual did not specify a frequency response at all. And sure enough, when I located a DR60 manual, the Specifications section indeed omitted any line for frequency response.

I also remembered reading that the DR60 had a low sampling rate, of either 6 kHz or 8 kHz. This is very low, since it means that the highest frequency that can be represented in the sampled signal (the Nyquist frequency) would only be 3 kHz or 4 kHz, which is only around the upper frequency of the old telephone system, and it made me question whether those references were correct (especially as one had said it was six samples per second, rather than six thousand). However, if the QR240 had a frequency response that only extended to 5 kHz, then it implied that its sampling frequency was likely only 10 kHz or so. The rates quoted for the DR60 started to seem more believable.

Googling around for as many relevant terms as possible, I stumbled on a remarkable document: what appeared to be a full technical manual and repair guide for the DR60! The document was linked from a huge list of technical documents on the site transkommunikation.ch, which is a German page effectively investigating EVP. The English in the manual is not perfect (but fairly standard for that era). The technical details in it, however, remove any doubt that it is a genuine technical repair manual.

Various details in the manual make it clear that the sampling rate of the DR60 is indeed 6 kHz.

However, I was amazed to see the following subsection of Section 3: Troubleshooting Guide:

From the technical manual for the DR60

From the Troubleshooting section of the technical manual for the DR60

Note symptoms (3) and (4): honking noises or erased sound is played back! This could easily describe what we heard on the DR60’s recordings. Now note the cause: “Formats in FLASH MEMORY are not linked properly.” What are we to make of this?

Reading through this section, it is clear, firstly, that the recordings are stored in the 2 MiB flash memory in one-second chunks. (The previous section, although a little difficult to decipher, together with the timing diagrams, seem to imply that the ADC is operating at a sample rate of 6 kHz, and that every 160 such samples are compressed using the CELP algorithm into 14 bytes of encoded audio, which equates to 525 bytes per second, or about 1.85 MiB of memory for an hour of recording.) These recordings are linked together in the flash memory, i.e. each single recording does not necessarily all lie in one contiguous block of memory, since if a number of non-sequential short recordings are deleted (there can be up to 99 different recordings stored on the device), then the memory will become fragmented, and not all of the available space would be available for a single long recording, if it was forced to be in a contiguous chunk of memory. This is not all that different to how memory in a computer or on a magnetic or solid state drive is frequently handled, although the format used for the DR60 is likely far simpler than that used in those cases.

The second key point is that this linked-memory format can become corrupted. The “honking” noise once per second likely means that the given audio track is wrongly linking to one-second chunks of encoded audio that are unrelated to the actual original recording, and may not even represent self-consistent encodings. Likewise, the playing back of “erased sound” suggests that the given audio track might be wrongly linked to a section of memory that has been marked as freed (i.e. its corresponding audio track erased). Furthermore, the symptom of total recording time being far less than the 60 minutes that should be available (minus a maximum of one second per recording, which only comes to one minute and 39 seconds maximum) suggests that in such cases the metadata keeping track of the layout of the linked sections in memory has become corrupted so that when those available sections are added up, the total falls short of the proper 60 minutes of recording time. Finally, a “Sorry” appearing on the unit when replacing the batteries suggests that the metadata can get so corrupted that the unit cannot even make sense of it at all.

None of this is strange: the same thing can happen to a hard disk, if power is suddenly removed from it while it is writing an atomic block of metadata, or if the computer hard-crashes or is hard-reset at that time. Every operating system comes with disk-fixing utilities that will check the disk for such metadata errors and fix them. (Even on modern Macs, enough bugs exist that I find myself needing to run this on one machine or another every month or two, and sometimes it reports and fixes corrupted metadata, despite the redundancies built into the Mac disk systems.)

What could cause this sort of “crash” for the DR60? Well, power failure, of course! Whether the batteries happen to run flat, are removed, or happen to be knocked physically causing a momentary loss of connection, any such event while the metadata is being written to the flash memory can potentially cause this sort of corruption. The fact that the “Sorry” message appears on inserting batteries reinforces the likelihood that this corruption is happening because of the sudden loss of power. Elsewhere in the repair manual it is made clear that, in addition to the two AAA batteries that the user supplies, there is also a 3V Lithium battery inside the unit, for backup—this is likely what (according to the user manual) provides about three minutes’ of grace when replacing the AAA batteries. Presumably this battery was rechargeable, and was recharged by the main AAA batteries. But the user manual does not provide any instructions for replacing this backup battery—and, based on the schematics and assembly diagram, it looks unlikely that it could be replaced by the user anyway. Thus, it seems like the device was designed in such a way that it becomes far more susceptible to corruption of the flash memory once that backup battery inside the unit reaches its end of life—which, assuming it was on the order of five years or so, was likely a reasonable design choice in the 1990s, but does not bode well for devices that are now a quarter of a century old.

Does the user manual (not just the technical repair manual) also refer to this sort of failure of the device? In fact it does:

From the user manual for the DR60

From the user technical manual for the DR60

Leaving the unit unused for a long time would likely cause the AAA batteries to run down. Once that happens, the backup battery will try to take over. However, if that only has a few minutes of charge available, then that would soon drop as well. It seems remarkable that the flash memory could become corrupted when power drops in this way, even with the device off, but clearly it was enough of a possibility that it was included in the user manual.

Although the user manual does not describe all the other symptoms of corrupted flash memory, the instructions for fixing it are the same as in the repair manual: reinsert the batteries while holding the “MODE” and “PLAY/SEL/STOP” buttons. The repair manual makes clear that this reformats the flash memory, which of course erases all recordings, but also lays out clean metadata for the audio storage.

The role of VAS

I have not yet mentioned the fact that, during our EVP sessions, the DR60 was used in “VAS” mode: Voice Activated System. Many audio recorders have had this feature over the decades, as it saves storage (memory, tape, whatever) when no one is actually talking into the device. (Remember, these devices were originally intended for dictation!)

On reading the manuals for the DR60, I was perplexed by one thing: why were we using VAS mode, when the device actually has capacity for a full hour of recording? (That’s the “60” in “DR60.”) It might be useful (or crucial) if the device were left in, say, a room for an extended period of time. But our recordings were short—around a minute each—and any interesting recordings were played into and recorded on a cellphone for preservation. (There is a better way to transfer audio through the headphone jack, but this is what was done at the time.) It would be easy enough to erase some or all of the recordings on the device, if memory was actually running out.

Now that I had discovered the possibility that memory corruption was the source of the anomalous recordings, I asked myself the question: could the use of VAS actually exacerbate the problem?

Although I don’t have a device to test this hypothesis, I believe that it may well be the case. Consider that, according to the repair manual, every fragment of audio must use at least one second’s worth of memory, which suggests that it is stored in one-second chunks. Having recordings stop and start frequently, because of VAS, will create far more fragments of audio that must be “stitched together” to create each full recording. If the metadata for linking these chunks is corrupted, then having many chunks provides far more opportunities for linking to random or previously-erased sections of memory. That, in turn, provides more opportunities for anomalous audio to be linked into the recording, which is then interpreted as EVP.

I now had a new question to ponder: Is this why VAS on the DR60 was used for EVP—to get more anomalous recordings? Is this why we obtained EVP on essentially every recording—even multiple on single recordings? I am speculating, but it is quite possible that it is. Note that I am not casting aspersions on any paranormal investigators making use of this mode on the device: it is only natural that the most productive settings will be reinforced and more frequently used, over time, as practitioners gain experience with different settings.

How do we test this hypothesis?

Clearly, the best way to test all of the above is with a DR60 device itself. However, if memory corruption is a possible explanation, then we now have a new variable to consider: how corrupted is the memory on any given particular device? It may well be that two apparently identical devices will give totally different results, under the same conditions (and I have read reports that this is indeed the case, when two DR60s are used side by side).

It therefore seems necessary to test a device that is reliably producing EVP. This may be more easily said than done: the most obvious test to perform, after confirming that the device is recording anomalous audio, is to reset it according to the manual’s instructions above. I am not sure that any paranormal investigator in possession of one of these devices—now worth thousands of dollars—is going to be willing to potentially destroy its EVP-producing abilities. (Presumably, it would be possible to make it return, by trying to crash the memory by yanking out the batteries repeatedly, but that might be a slow, trial and error process.)

On the other hand, this is the only honest way of trying to assess if this flaw of the DR60—described in the manufacturer’s own manuals—may be responsible for these anomalous recordings. (Of note, the Troubleshooting section of the QR240’s user manual does not contain any similar instructions for resetting this device; presumably, Panasonic’s engineers had fixed the flaw that plagued the DR60 by the time that they manfactured the QR series.)

A simpler, less “destructive” test would be the following: find a DR60 exhibiting good EVP, transfer all recordings to some other device, erase all recordings (a simple one-step process), and check whether the available recording time reported is 60 minutes. If not, then this would be immediate proof positive that the device has corrupted flash memory. (Showing 60 minutes, however, would not be proof against corruption: only resetting the device can ensure that.)

Does this mean that we are settling on Option (1)? What does that imply?

If, after testing an actual DR60, it turns out that the above hypothesis turns out to be feasible—that memory corruption explains anomalous recordings in the DR60—then, unfortunately, it would mean that Option (1) has been established.

Again unfortunately, that would imply that any recording made by the DR60 may well just be the product of random corruption. Such recordings would not be evidence of EVP.

Does anyone near Washington DC have a DR60 I could test?

Given the above, I know that the likelihood of this is very low. However, after having had this website running for 28 years, I know that, given enough time, the right people will almost always stumble across one of my pages, eventually. Let me know if that person is you!

Update: Halloween 2022

Over Halloween 2022, Sally and I were able to again meet up with the person who had first demonstrated the DR60 to us—indeed, they and their colleagues had three DR60s between them. We got to witness numerous EVP sessions, at length, many with two DR60s sitting side by side. We also got to chat extensively with the person about the DR60. And I was also able to (quickly) test the DR60 with both a test audio sweep file, as well as with the ultrasonic generator above. Some things we established:

  1. The person does, indeed, reset their DR60 frequently, and EVPs still seem to appear almost immediately. (So the flat backup battery theory seems to not be the explanation—or not the full one, anyway.)
  2. The person had not seen the “Sorry” message on their DR60, but they had seen it on other people’s DR60s.
  3. When their DR60 starts to exhibit the “honking once per second” behavior, they know that it is time to reset it.
  4. We have not yet determined if the total recording time showing is incorrect.
  5. The DR60 shows aliasing for high audio frequencies, but the amplitude is probably not sufficient to explain all of the anomalous recordings.
  6. The DR60 does not respond to the ultrasonic sounds, but, interestingly, during the final sweep test signal, the VAS stayed on, and exhibited the same sort of “staticky” sounds heard on EVP recordings.
  7. Critically, when two DR60s are placed side by side, things that were heard in the room at the time of recording are audible on both recorders, but things that were not heard in the room at the time only appear on one or the other of the recorders.
  8. When the person accidentally left the DR60 recording after the EVP session had ended—so that the VAS continually stayed on, because they were explaining things to us—no EVPs were present in that part of the recording.

Observation (7) here establishes that the EVPs picked up by the DR60 are not measurements of a physical pheneomenon, since any such physical phenomenon should affect two DR60s equally, whether through Option (2) or Option (3) above, or indeed any other physical explanation that I hadn’t thought of.

That leaves Options (1) and (4).

Observation (8) strongly suggests that Option (1) is the true explanation, and that the root cause is most likely the software controlling the VAS system. Recall that the VAS needs to “stitch together” stored pieces of audio, and that this storage of data occurs in one-second chunks. If there was some software bug that doesn’t properly handle this “stitching”—say, by including random data from the rest of the second—then the CELP compression method used in this recorder will take that random data and turn it into something that sounds somewhat like speech.