Overview and History of the Acoustical Evidence in the Kennedy Assassination Case
by D.B. Thomas
CHALLENGES TO THE ACOUSTICAL EVIDENCE
The first challenge to the HSCA acoustical analysis came from the FBI laboratory. It was published in two installments in volume 52 of the Law Enforcement Bulletin by Special Agent Bruce Koenig. It is Exhibit A in a long list of examples of how the FBI laboratory had prostituted itself in pursuit of convictions rather than justice. Koenig, a voice print expert, had little grasp of basic acoustical science. In his criticism of BBN he wrote,
"BBN eliminated a number of possibly useful impulsive patterns . . .Four impulsive patterns were eliminated because the specified motorcycle would probably be traveling too fast to be in the motorcade. However, the impulse could have been received by another motorcycle in the motorcade with an open microphone or in another part of the city. In other words, six other gunshots may have occurred in Dealey Plaza, according to the BBN analysis ..." Koenig in FBI Law Enforcement Bulletin 52(12) pt 2, p.5
But nothing resembling this fantasy can be found in the BBN report. BBN had identified five suspect patterns plus one attenuated pattern in their laboratory examination of the dictabelt. The latter was subsequently eliminated because it failed to match to any of the test shots from Dealey Plaza, not because of anything to do with motorcycle speed.
In his criticism of the sonar analysis, Koenig wrote simply, “It is not known which characteristic Weiss and Aschkenasy actually used in their analysis.” Ibid pt. 1, p. 8.
His own statement pretty much summed up Koenig’s problem. He truly did not know what Weiss and Aschkenasy had done. For the record, the characteristic used by Weiss and Aschkenasy in their analysis was echo delay time. Bottom line is that the FBI analysis was hardly a serious challenge to the acoustical evidence, and thus does little to strengthen the case. Unfortunately, most of the challenges to the acoustical evidence that one finds on the internet are of this sort: long on sophistry, short on science.
Court TV – Forensic Files
The only other challenge to the core acoustical evidence was featured on the cable television program, Forensic Files. The show’s producers had contracted with a firm called Sensimetrics. During the program an expert named Robert Berkowitz asserted that the suspect sound on the DPD recording alleged to be the grassy knoll shot, did not in fact match to the test shot from the grassy knoll.
Because I had been interviewed for the show, and many of my graphics were used by the program, the producers sent me a copy of the Sensimetrics report. Their error was immediately obvious. Because the parameter being compared is the echo delay time, it is essential that playback speed of the evidence recording be the same as the original recording speed. In their comparison, Sensimetrics had used the wrong tape speed. The source of the error traces back to the NRC report. The original recording was made on a Dictaphone machine. The DPD playback used in the HSCA analysis was a magnetic tape recording made by playing back the dictabelt on a rented Dictaphone machine. Because these machines were designed for dictation, the playback speed had a variable control making it inevitable that the playback speed would not exactly match the original recording speed. Using the 60 hz motor hum from the original machine, BBN determined that the police recording was made at a playback that was around 5% too fast. But instead of slowing the CH-1 recording, Sensimetrics had sped it up by 5 percent. This is the same mistake made by the NRC panel in their report and may be the origin of Sensimetrics mistake. When I saw the error in the report I sent an e-mail to Robert Berkowitz at Sensimetrics asking for clarification on why he had sped up the recording instead of slowing it down. Without a reply, Berkowitz sent me a disc with his computer program so I could perform the comparison myself.
Screenshot of Sensimetrics Impulse Analysis program.
(view larger version)
It turned out that the playback used by Sensimetrics was not the Dallas police recording from 1963, but a playback made by the FBI and NRC panel in 1982 with 60 Hz speed correction built in. Without any correction for speed, the sensimetrics program demonstrated that the test shot and the suspect pattern do match. Just to be sure I notified my colleague Michael O’Dell who confirmed that the recording was the 1982 NRC playback and that the patterns do match. Anyone, can download the Sensimetrics program from the Court-TV website and confirm the match for oneself. Just set the speed adjustment to the zero setting.
The National Research Council
Most of the challenges to the acoustical evidence do not involve the core evidence, but rather the corroborative evidence. The foremost example is the NRC report published in 1982. This report is remarkable in several respects. In spite of the fact that the panel who issued the report were all highly respected, internationally renowned scientists, including two Nobel Prize winners, it is one of the most unscientific reports that I have ever had the displeasure to read. It is replete with factual error, most of which is prejudicial to the HSCA’s expert analysis. The introductory section begins with this distortion on page 9 of the report:
"The BRSW [BBN] correlation was meant as a selective screen. Only candidate shots that gave a binary correlation greater than 0.6 were studied further. In all, fifteen pairs, involving only six sets of impulses on the Channel I recording, survived this screening."
Everything in this statement is wrong. The BBN correlation was not a selective screen. The screening was done in the laboratory using acoustical criteria, which is not explained in the NRC report. Five sequences were identified as candidates by the laboratory screening because they had the basic acoustical signature of rifle fire. The binary correlation procedure was applied in the field test and was the last acoustical test applied, nothing was studied further. The second sentence says that “only” six sets of impulses survived the screening. In the binary correlation test five (5), not six sets of impulses, passed the test. The way the sentence is constructed it implies that there were others as many as 15 impulse sets that did not survive the screening. The NRC report is easily misread to imply that there were fifteen candidate sets of impulses on the recording and only six passed the binary correlation test. These weasel words are undoubtedly the source of the error on this point made by Koenig in his FBI report.
On page 10 of the NRC report it says,
"Of the six sets of impulses that give high binary correlation coefficients with test shots, BRSW selected four as likely assassination shots by eliminating those whose echoes were inconsistent with a reasonable trajectory." [ NRC Report p. 10.]
Everything in this statement is wrong. First of all, five, not six sets of impulses gave high binary correlation coefficients. Secondly, no suspect sets of impulses were eliminated because they were inconsistent with a reasonable trajectory. Five suspect patterns were identified in the lab screening and all five passed the field test (Binary correlation). Importantly, and contrary to the NRC report, all five fall in alignment with a reasonable trajectory. None were eliminated because they had an unreasonable trajectory – never happened. The author of the introductory section of the NRC report, apparently Luis Alvarez (although all co-authors bear responsibility), deliberately misrepresented the acoustics study to make it appear that the results were equivocal and that the analysts had cherry-picked the data to reach an arbitrary conclusion. The NRC panel spent two years trying to find a flaw in the acoustical analysis. Unable to find a serious flaw, they resorted to distortion and misrepresentation. Luis Alvarez was asked to chair the NRC panel because he was a long time defender of the Warren Commission. Alvarez had every right to defend his positions on this issue. But he had no business serving on a tax-payer funded panel whose responsibility was to provide an unbiased review of the acoustical evidence.
One of the important corroborations to the acoustical analysis was the fact that a second laboratory had confirmed the work of the first. The critical finding by BBN was that there was a gunshot from the grassy knoll. The sonar experts had expanded the analysis and confirmed that finding. But the NRC report claimed otherwise.
"The identification of shots and impulses by BRSW was completely different from that by WA as demonstrated by the more than 200 millisecond (or more than 200 ft.) displacement between the two identifications...the BRSW analysis missed the identification that WA considers to be the primary one." [ NRC Report p. 13.]
In a separate article written by one panel member, Richard Garwin, it states,
"Both sets of workers could not be correct." Garwin (198 ) p. 205.
But both sets of workers were correct. The statement in the NRC report is false. The team at BBN and WA had identified the same pattern. The difference was that BBN’s test had involved only the largest impulses whereas WA in expanding the analysis had tested the entire pattern including the smaller impulses. This was because the two groups used different thresholds to separate putative gunshot noises from the background noise.
Note that in the test shot pattern there are two clusters of impulses. Because Weiss and Aschkenasy’s study involved associating the echoes in the test shot with specific structures in Dealey Plaza, we know that the first set of impulses originate with structures on Elm Street. The second cluster of impulses originate with the structures on Houston street. There is a 200 millisecond gap between the two clusters. That is what causes the 200 ft displacement mentioned in the NRC report.
The approach by the acoustical experts was a classic example of trying to separate noise from signal. When BBN conducted their comparison they used a noise threshold based on the average noise level of the motorcycle motor over the ten second interval of the recording that includes all five of the suspect impulse patterns. Giving a close look at the suspect pattern one can note that if this noise threshold line is lowered slightly, it includes the second cluster, but not the first. That is why the BBN test identified this second cluster as a likely gunshot echo pattern. The BBN test was blind in that they did not know which impulse patterns might pass their tests let alone where the sounds might have originated if they were gunshots. Weiss and Achkenasy knew these things. Weiss and Aschkenasy used a threshold based on the motor noise during the one sec segment immediately anteceding the suspect impulses. This analysis included both clusters and demonstrated with the fuller comparison that the pattern still matches to the test shot. Actually, W&A used three different thresholds and obtained significant matches at each level. The threshold shown here is the most appropriate one as explained in their report. Thus, both workers had identified the same pattern, not completely different patterns as the NRC report falsely claimed. It is a testament to the robustness of the BBN test that they were able to correctly detect the grassy knoll shot though using only a smaller set of impulses. Yet the NRC exploited the difference in order to discredit the analysis. Make no mistake about this, this misrepresentation is deliberate. Dr. James Barger the lead scientist sat down with this panel and explained to them exactly why there was a 200 msec difference, just as I have explained it here. These errors are not typo’s or inconsequential lapses, nor do they involve tangential issues. These gross errors misrepresent the core acoustical evidence itself.
These were far from the only errors in the NRC report. The statistical analysis was fudged, a subject I dealt with in my 2001 paper in Science & Justice. The appendix of the NRC report contains conclusionary statements that are unattended by analysis or evidence. One example being the claim that the presence of sirens on the recording two minutes after the assassination is somehow problematic to the acoustical evidence with no explanation offered as to how or why. The NRC panel members were acting as advocates for the official mythology, not as objective scientists.
The NRC Redux
The NRC’s conclusion that the acoustical evidence was invalid, is based on the assertion that the sounds identified as gunfire were not synchronous with the time of the shooting. On the day that Kennedy was assassinated, the Dallas police were using two radio channels. Ch-1 was used for routine police communications. An auxiliary frequency, Ch-2 was used for special events, in this case, for the police escort of the president’s motorcade. Events on the separate channels can be synchronized because there are simulcasts, that is, broadcasts that are common to both channels. One is a deliberate simulcast by the dispatcher which begins with the phrase, “Attention all emergency vehicles…”. The dispatcher had a switch on his console that allowed him to make simulcasts. But most of the simulcasts were accidental due to a phenomenon called crosstalk. Basically, if two police units are close together, but tuned to opposite channels, and one opens a microphone, it can capture broadcasts from one channel and instantly simulcast them over the other. This happened four times just during the 5-1/2 min sequence when the motorcycle microphone was open.
The NRC panel’s synchronization was based on the juxtaposition of the suspect gunfire to one of the simulcasts, a crosstalk of a broadcast on Ch-2 by Sheriff Decker about a minute after the assassination, ordering his men to go to the grassy knoll area and, “Hold everything secure.” The same broadcast occurs on Ch-1 at the end of the sequence of sounds identified as the gunshots. The NRC panel concluded that the juxtaposition of the sounds to a broadcast a full minute after the shooting was proof that the sounds, whatever they are, cannot be the assassination gunfire. The NRC redux (Linsker et al., 2006) repeats this assertion. The problem, as I pointed out in my 2001 paper, is that it depends on which instance of crosstalk one chooses as the tiepoint between the channels. A deliberate error of omission on the part of the NRC and the NRC redux was the failure to consider the fact that there are five instances of crosstalk and that in every single instance, the time between them is different on the two channels (Table 3).
Table 3.- Speed corrected playback time intervals between simulcasts on the Dallas Police recordings (in seconds). Data from O’Dell (2003)
|CHECK (1) to HOLD||10||99||89|
|HOLD (2) to YOU||174||143||31|
|YOU (3) TO ALL (4)||15||12||3|
|YOU TO ATTENTION (5)||114||90||24|
1 = "I’ll check it", 2 = "Hold everything secure", 3 = "You want me to hold traffic on Stemmons", 4 = "I’ll check all these motorcycle radios", 5 = "Attention all emergency vehicles."
Very simply, the data in this table proves that playback time on these recordings is not real time. One of the problems is the sound actuation function which stopped the recorders during dead air. In theory, this should not have been a problem on Ch-1 because the constant motorcycle noise would have kept the recorder running. But this was clearly not the only problem. Consider the 3 sec displacement between the two consecutive simulcasts from Sergeant Bellah who was searching for the motorcycle with the stuck microphone. If this discrepancy had been caused by recorder stoppage, then there should be four sec of dead air between the two broadcasts, but there isn’t. But inasmuch as the rotation time of the audograph disc is around 3 sec, the simplest explanation is that the stylus skipped a groove at this point. Because both recorders were stylus in groove arrangements, they were prone to displacements, a problem which is very obvious on Ch-2 which is studded with repeats. Moreover, if it is true that the recordings are not the originals, but only copies, then the possibilities of artifacts causing offsets in time multiplies. Regardless of what has caused the time offsets documented in Table 3, the fact is that the juxtaposition of events to the crosstalks is not a reliable indicator of synchronization, or lack thereof, because even the crosstalks themselves do not synchronize with one another. Neither NRC report reveals, let alone deals, with the data in Table 3, because it directly contravenes the basis for their position.
Jack Weaver photo, showing lead car about 120-150 feet ahead of the Presidential limousine.
(view larger version)
It should be obvious however, that whatever phenomena is imposing the time offsets on the recordings, that the amount and likelihood of an imposition of a time offset between any particular recorded event and a corresponding crosstalk will be the amount of time between them. The farther apart they are, the greater the possibility that an offset has been imposed. Hence, if one is going to use the crosstalks to synchronize events, one should use the crosstalk closest to the incident in question. The NRC panel knew this (because I pointed it out to them) which led to their next major error of omission. In spite of their argument that the suspect sounds are not synchronous with the assassination, at no point in their report do they ever identify the actual time of the assassination. The actual time of the assassination can be fixed by the context of the broadcasts from the motorcade on Ch-2. These broadcasts emanated from Police Chief Jesse Curry who was with the first car in the motorcade. The Weaver photo shows the motorcade in Dealey Plaza with the lead car, in accordance with Secret Service rules, about 120 to 150 ft ahead of the President’s limousine. The transcript of the Ch-2 broadcasts (Table 1) shows that the last broadcast by Curry just 20 sec before the crucial “Go to the Hospital” broadcast, was an announcement that he was at or approaching “… the triple underpass.” The triple underpass is the railroad bridge at the western edge of Dealey Plaza. When the lead car was at or near the underpass the presidential limousine must have been in the mid-section of Elm Street, the position where the shooting occurred. In his September 2003 account for the oral history project at the Sixth Floor Museum in Dealey Plaza, Winston Lawson, the Secret Service Agent who was in the lead car, recalled that he heard the shots and recognized them as gunfire, just as the lead car was arriving at the Triple Underpass. Hence, Curry’s broadcast that he was at the triple underpass is a marker for the time of the assassination.
The crosstalk closest to this event is the broadcast just two sec earlier by Deputy Chief Fisher saying the words, “Naw, that’s all right, I’ll check it.” The last three words, “I’ll check it” occur on Ch-1 just two sec before the first acoustically identified gunshot. Thus, the Fisher crosstalk establishes exact synchrony between the assassination and the acoustically identified gunfire. Neither the NRC report or the NRC redux acknowledges this fact. Instead, without telling the reader the significance of this broadcast, they deny that it is an instance of crosstalk. Because I pointed out the significance of this crosstalk to them, they cite me as the source of the claim that the two broadcasts are crosstalk and further cite this as a “serious error” on my part. In point of fact, the fisher broadcast is cited by the Dallas police officers who prepared the official transcripts as the exemplar of the crosstalk phenomenon, another inconvenient fact not mentioned by either NRC report.
Chart showing instances of crosstalk on the two channels.
At my suggestion the NRC redux performed a voice print comparison of the two corresponding broadcasts but withheld the result. They applied the test, a computer driven comparison of the frequency spectrum, to the Decker, Bellah and Fisher crosstalks. They provided the results for the first two comparisons, graphics showing the computer score of the correlation between the corresponding signals, but no such graphic is shown for the Fisher comparison, in spite of the fact that it is the only one that is at issue. Instead, they show a graphic representing what they refer to as “accidental peaks” evidently in an effort to downplay the significance of the results of the comparison. The results of the test are revealed only in a brief statement in the text which states that the peak so obtained was no larger than some of these “accidental” peaks. But the accidental peaks are the same amplitude as the peaks obtained with the comparison of the Decker and Bellah crosstalks. The implication is that the critical peak must also be accidental, or as scientists describe such instances, a false positive. That is certainly a valid explanation for the result. If one were to apply the test to two separate utterances of this phrase by Fisher, or by someone with a similar voice pattern, the results would likely be the same. But how likely is it that someone else said the same phrase on Ch-1 and it is just a diabolical coincidence that it happened very close to the suspect sounds? When I questioned Richard Garwin, a co-author of the report by Linsker et al, he claimed that the phrase occurs frequently in the broadcasts. In point of fact, the phrase occurs two times, once on Ch-1, just before the suspect sounds, and once on Ch-2, just before the assassinaton.
In order to explain away the result of the voice print comparison, Linsker et al. offered a number of non-sequiturs; the most inane being the argument that it can’t be crosstalk because it’s out of synch with the other crosstalks. They also argued that the peak achieved was with not at the expected speed. But in this context the “expected” speed is meaningless. The expected speed is determined from the 60 Hz motor hum. But it is known that the actual speed of the recording mechanism at any given instant wobbles around the motor speed. It is precisely for this reason that the computer makes the comparison iteratively at increments of speed warp deviating from the expected speed. In a classic example of circular reasoning, when the test was applied to the crosstalk instances favored by them, the “correct” speed was defined as the speed that generated a peak! Furthermore, a comparison at the incorrect speed would not generate a peak. A comparison at the wrong speed can generate a false negative, but it cannot cause a false positive. A peak can only indicate that the frequency spectrum is similar to a significant degree.
Relevantly, the IBM lab had a test which can separate a false positive peak, from a true positive. They called this test the d-warp test. The “d” stands for duration. The peak obtained in the comparison of the dispatcher’s broadcasts would be expected because the words and vocal quality measured in the frequency spectrum, are the same. But the duration of the transmission would be slightly different just because they are different utterances, whereas the duration would be exactly the same if they were true crosstalk. The IBM group applied the d-warp test to the HOLD and CHECK broadcasts demonstrating that at the speed warp which generated the peak, the duration was also the same. Conversely, at the speed warp which generated the peak with the dispatchers “ten-four” broadcasts the duration was different, demonstrating the principle and the power of the test. The IBM report states (on p. 222) that the peak from the comparison of the CHECK broadcasts had failed the d-warp test. But again, the results were withheld. Only on closer examination does it become clear that the d-warp test was applied not to the peak obtained at the correct delay (but “unexpected” speed warp), but to one of the accidental peaks obtained at the expected speed warp, but not at the expected delay. By this sleight of hand they seek to convince the reader that the CHECK broadcast is not a crosstalk. One might suppose that the reports authors really did apply the d-warp to the peak at the correct delay, as they certainly should have, but the result is not in their article. It is difficult for this reader to believe that they would have withheld a result which supported their conclusion.
By no means has all of the evidence on the DPD tapes been exhausted. Here are some examples of further analysis that could be done, and of course, the results of further studies almost always suggest even further avenues that might be explored. Unfortunately, this fact provides ammunition to some critics who argue that therefore the analysis is incomplete and more studies should be done before any conclusion is accepted. As Socrates once said, the partisan has no respect for the rights of the argument.
1. The sonar analysis applied to the grassy knoll shot should be applied to the other shots, in particular the “rogue” shot, the second pattern in the sequence, to confirm or refute that it is a shot and to determine its origin. Applying the same analysis to the book depository shots serves as a check on the degree of precision that the analysis can produce. Ideally, this test would include firing test shots from an array of potential sniper locations in the buildings near the intersection of Elm and Houston.
2. BBN scientists suggested that a further study of the power line hums might help resolve the issue of whether the dictabelt is the original or a copy. If the dictabelt is the original, then a discontinuous secondary power line hum might also help resolve the problem of the displacement of the cross-talks. If a stylus displacement was involved, the re-recorded section of the belt might have a detectable secondary hum.
3. All of the playbacks of the Ch-2 audograph disc recording, including the FBI/NRC playback, contain skips and repeats. If the skips and repeats could be eliminated (a laser study of the acoustical grooves might resolve this problem) an accurate timeline of events might be achievable. Such a study would be more valuable than the presently proposed laser study of the Dictabelt.
4. An exacting photogrammetric analysis of the various newsreels could be used to refine our understanding of the position of the various vehicles, including the police motorcycles, at or near the time of the shooting. As it stands, the evidence which tends to show that Officer McLain was in the right position as dictated by the acoustics may or may not be strengthened. The Wiegman and Dorman films seem to be key among this evidence.
5. The broadcasts recorded on the police tapes tend to incriminate elements of the Dallas Police as accessories after the fact in the assassination. This evidence includes not just the DPD but also the Sheriff’s department recordings. The audio evidence has been under-exploited, especially when compared to the video evidence. Among the issues are why did the radio dispatcher instruct officer J.D. Tippit to leave his beat and go to Lee Harvey Oswald’s neighborhood in central Oak Cliff long before there was evidence that Oswald was involved in the assassination. Why was there a focus on a station wagon with armed suspects in that neighborhood immediately after Tippit was shot? Why did Oswald ask a taxi to take him to a Mexican restaurant in Oak Cliff after the assassination. And most importantly, were Tippit’s friends at the restaurant as part of an escape plan?
Zapruder film frame 175, where the acoustics evidence places the first shot.
1. Circumstantial evidence indicates that the motorcycle with the open microphone was part of the police escort with the President’s motorcade.
2. Half way through the motorcycle segment, at approximately 12:30 local time, there are five sound patterns with the acoustical characteristics of rifle fire.
3. All five of these sound patterns match to a significant degree to the echo pattern of test rifle shots fired in Dealey Plaza.
4. The sequence of gunshots indicated by the acoustical evidence is three shots close together, each about a second apart and then two more shots a split second apart about five seconds later. The actual time gaps between the shots are: 1.6, 1.1, 4.8 and 0.7 sec.
Picture taken from behind the grassy knoll fence.
5. The fourth sound pattern in the sequence matched to a test shot from the grassy knoll. The others matched to test shots fired from the TSBD. However, the second suspect sound in the sequence occurs to close to the suspect sounds on either side to have originated with shots from the suspect weapon found in the TSBD because tests with this weapon established that it could not be cycled (fired and refired) in less than 2-1/4 seconds (reports that the weapon could be cycled in 1.6 sec are fallacious).
6. The wounding sequence seen in the Zapruder film, with Governor Connally struck at Z-224, and President Kennedy struck at Z-313, knocking him backwards, indicates a shot from the book depository followed by a shot from the grassy knoll, 4.8 seconds apart. This video sequence precisely matches the audio sequence.
7. Blur analysis of the Zapruder film shows that Zapruder was startled causing the camera body to jerk at Z-227 and at Z-313. Factoring in bullet speed for impacts at Z-224 and Z-313, and the speed of sound to produce a muzzle-blast induced startle in Zapruder at Z-227 and Z-313 supports wounding shots originating respectively at the TSBD and the grassy knoll.
Rifleman's view from behind the grassy knoll fence, at the spot where acoustics evidence indicates the fatal shot was fired.
8. Newsreels showing the motorcade in Dealey Plaza indicates that a motorcycle ridden by officer H.B. McLain was in positions before and after the shooting such that with a reasonable trajectory he could have been in the acoustically predicted locations to have recorded the gunfire at the time of the shooting. No other motorcycle unit is in a position such that a reasonable trajectory would have been at the predicted locations. In testimony to the HSCA McLain recalled that his radio microphone tended to stick in the on position.
9. The police were using two radio frequencies at the time of the assassination. Ch-1 contains the suspect sounds identified as gunshots. Ch-2 contains broadcasts from the President’s motorcade. The juxtaposition of the suspect sounds to a simulcast of the words, “I’ll check it”, a broadcast found on both channels, establishes exact synchrony between the suspect sounds and the actual time of the assassination. That is, the suspect gunshot sounds were deposited on the police recording at the exact instant that President Kennedy was being assassinated.
Barger JE, Robinson SP, Schmidt EC & Wolf JJ. Analysis of recorded sounds relating to the assassination of President John F. Kennedy. US Congress, House of Representatives, House Select Committee on Assassinations Proceedings vol. 8. 1979.
Bowles JC. The Kennedy assassination tapes: a rebuttal to the acoustical evidence theory. Pp. 313-410, In Savage G, JFK First Day Evidence. Shoppe Press, Monroe LA. 1993.
Linsker R, Garwin RL, Chernoff H, Horowitz P and Ramsey NF. Synchronization of the acoustic evidence in the assassination of President Kennedy. Science & Justice 45: 207-226. 2005.
O’Dell M. The acoustic evidence in the Kennedy assassination. 2003.
Ramsey NF, Alvarez LW, Chernoff H, Dicke RH, Elkind JI, Feggeler JC, Garwin RL, Horowitz P, Johnson A, Phinney RA, Rader C, and Sarles FW. Report of the Committee on Ballistic Acoustics. National Research Council (USA) Prepared for Department of Justice, Washington D.C. Report No. PB83-218461. 1983.
Report of the President’s Commission on the Assassination of President John F. Kennedy, U.S. Gov. Print. Off. Wash. D.C. 1964.
Thomas DB. 2001. Echo correlation analysis and the acoustic evidence in the Kennedy assassination revisited. Science & Justice 41:21-32. Errata. Science & Justice 41:132. 2001.
Weiss, MR & Ashkenasy, A. An analysis of recorded sounds relating to the assassination of President John F. Kennedy. U.S. Congress, House of Representatives, House Select Committee on Assassinations Proceedings vol. 8. 1979.