Eight subjects were recorded speaking texts in synchrony in groups of 1, 2, 4, 6 and 8 speakers at a time. The question asked was whether either speech timing or inter-speaker synchronisation would be manifestly di erent as more speakers speak together. Results showed no appreciable changes to overall speech timing as number of speakers varied. Examination of asynchrony scores likewise showed no e ect of number of speakers, even as texts from very di erent genres were used. The results suggest that synchronous speaking in a laboratory is importantly di erent from joint speech found in ritual, protest, and elsewhere.
Joint speech is found wherever multiple people utter the same words at the
same time [
One experimental avenue that has produced some limited insights is
provided by the vehicle of synchronous speech [
Past work on synchronous speaking has used dyads (pairs of speakers)
exclusively. The body of ndings includes the observation that speakers can
synchronise e ectively and without practice [
When joint speech is encountered, there are typically very many more than two speakers involved. Synchronisation is much more lax than found in these dyadic synchronous speech experiments, and, of course, the motivation and context are radically di erent. In what follows, a small experiment is reported that employs the synchronous speech methodology, but extends it to more speakers, allowing a comparison of the temporal alignment of 2, 4, 6 and even 8 speakers at a time. We pose two substantial research questions. Will the addition of more speakers lead to systematic changes in the overall temporal patterning of speech? And will the addition of more speakers lead to less overall synchrony within the group? This will allow us to better understand the di erence in group performance between the experimental situation of synchronous speaking and the collective behaviour of joint speaking in the wild. We know that prosody in chanting is often substantially altered, but this may be a function of repetition, which is frequently present, or of associated body movements, such as st pumping, or of many other potential causes. We do not know if it is a necessary e ect of group synchronisation, though observation of joint speech done instrumentally, such as in public swearing of an oath of allegiance, typically exhibits less prosodic stylisation than well practiced pledges, chants, and prayers. We also know that synchrony among speakers in ritual and protest is typically not as tight as that found under laboratory conditions of synchronous speaking. 2
Eight speakers took part (age range: 21{46, 2 Female, 6 Male). All were student volunteers and native speakers of English. No reward was given for participation and all experimental procedures were approved by the UCD research ethics committee. Recording was done in a single class room (approx 3 m x 4 m) and subjects each wore head mounted microphones to minimise cross-speaker contamination. Each subject was recorded on a spearate channel. An ambient recording of all speakers was also captured with a separate microphone.
Five texts were prepared that spanned genres. They included one regular and one irregular wordlist, the Hail Mary prayer, a short prose piece with irregular stressed syllable onsets, and a metrically complex poem. All texts had been used in previous synchronous speech experiments. There were 40 trials. On each trial, 1, 2, 4, or 6 subjects were randomly selected, or all 8 subjects took part, and the 5 texts were read in randomized order. This provided eight complete recordings of all texts for each xed number of speakers. Speaking began on a signal from the experimenter. No tempo instructions or other guidance was given.
There are several ways to estimate synchrony across any given pair of speakers
[
Initial results will be presented here for the rst time. A rst question to be asked is whether overall speech timing is a ected by the constraint of speaking in groups of di erent size. To pursue this issue, we rst examine the intervals between stressed syllable onsets in the two wordlists. The regular word list consisted of eight trochees (Borrow, Dancer, Butter, Dagger, Boiler, Doggie, Body, Deeper ) while the irregular wordlist had a similar strong-weak alternation, but with an irregular patterning of word edges (Debug, Debasing, Ball, Degree, Bandana, Beach, Beginning, Duck ).
Fig. 1 shows interval durations for the two wordlists as a function of
position in list (x-axis) and speaker numbers. Each list displays idiosyncratic timing
properties that are highly preserved across changes to speaker number, and,
crucially, there is no evidence here for any systematic change as speaker numbers
vary from 1 to 8. The alternating pattern of relatively shorter and longer
intervals is of a kind with previously presented results from both speech and typing
[
We repeat the analysis now with a much more rhythmically complex text. The poem \Kill a cat" displays a binary rhythm in lines 1, 2, and 5, but switches to a triple metre in lines 3 and 41. The text is:
Kill a cat, kill a cat Bash its brains in with a bat.
Their nine lives expire When tossed in a re, so Kill a cat today. 1 Apologies to cat lovers everywhere. N=1
N=8 ) c e s i(no .06 tr a u D ) c e s i(no .06 tr a u D are computed on a pairwise basis, there are many more pairwise comparisons when n = 8 than when n = 2 (224 vs. 8).
From the gure it is clear that asynchrony remains more or less constant as the number of speakers increases, irrespective of text type. There is some evidence that the simpler texts (word lists) result in slightly tighter synchrony than the complex poem or prose passages, but the principal result here is a lack of e ect of the number of speakers. 4
Joint speech and synchronous speech are not the same thing. The rst is found in
a wide variety of situations in which collective purposes, passions, and aspirations
are made manifest, and in which di erent kinds of identies are brought forth or
enacted, whether that be as Buddhist monks, concerned citizens, or Arsenal
supporters. The latter is found in the laboratory, and although there too people
speak in synchrony with one another, their purposes are starkly di erent, and
the speech produced has its own speci c characteristics. It usually exhibits a very
strong synchrony that is alien to joint speech in the wild. But the circumstances
under which joint speech is produced vary greatly and they frequently include
contextual factors and historical particulars that leave their stamp on the speech.
Thus prayers and ritual incantations are typically repeated many times over.
Protest chants are repeated and fequently augmented with st pumping, clapping
or noise making of various kinds. Football chants too typically involve repetition
and musical elements. (In studying joint speech, a determinate border between
speech and song or music can no longer be sustained [
A signi cant di erence between the context in which joint speech and synchronous speech occur is the restriction of all previous synchronous speech experiments to dyads, whereas joint speech typically involves many more speakers. The present work thus served to see whether this factor, number of speakers, can account for any of the observed di erences between the two kinds of speech. The result is in the negative, both when overall temporal patterning is examined, and when pairwise asynchronies are computed. Although a negative result, this is a nding of some signi cance, as it allows us now to more clearly distinguish between the two kinds of speaking, and to recognise that synchronous speaking is not free of its own contextual imprematur that cannot be attributed simply to the smaller number of participants.
One important lesson to be learned from this small experiment is that the study of joint speech demands a great deal of sensitivity to the context in which it occurs. In keeping with a general trend in phonetic and linguistic research, it is necessary to recognise that the laboratory is not a neutral space in which a pristine object of study, speech, can be found. It is, rather, itself a rich context imbued with its own characteristics that indelibly mark the speech produced therein. The enhanced synchrony found in synchronous speech, and the associated characteristic phenomenon of joint cessation when an error occurs, these are not a result of a lack of numbers, but a product of an altered situation of speaking.