Instrumentalizing Voice

Social and Technical Dimensions of Recorded Voice

In technical applications of voice in telephony interactive voice response services (IVR), the term “voice” usually refers to speech that has been recorded in order to fulfill specific, repetitive, answering functions for business and other organizations. IVR platforms have both software and hardware dimensions in which computer voice files (usually of the “ .Wav” type) are activated by software that responds to either Dual Tone Multi-Frequency (DTMF) “touch tone” keypad or, in the case of voice recognition systems, “open dialogue” prompts from the speech of the caller (Attwater, McGrail, & Sargent, 2000). Most IVR continues to be based on DTMF keypad inputs to initiate the menu choice the caller makes after the “welcoming prompt” at the start of the call. DTMF IVR voice prompts are very short, lasting only about 15 seconds at the welcoming prompt where the various menu choices are introduced and then the options that deliver information take around another 10 seconds or so each. For example, the following is the opening command prompt of a DIY service:

Thank you for calling B&Q. You now have three options: for store opening hours, please press 1. For directions, press 2. To speak to a member of staff, press 3. If you do not have a touch-tone telephone, please hold to speak to a member of staff. (B&Q London Call Line accessed 12.6.2012)

Selecting Option 1 brings the following message:

Our normal opening hours are as follows: Monday to Friday 7 a.m. to 9 p.m. Saturday 7 a.m. to 8 p.m. Sunday 10 to 4 p.m. We are open every bank holiday except Easter Sunday and Christmas Day. (B&Q London Call Line accessed 12.6.2012)

This message then terminates with the return option: “To hear these options again, please press star.” Following all the prompts for any one service menu therefore produced around 3 to 4 minutes of recorded voice data.

Recorded telephonic voices of this type are largely overlooked in the study of “secondary mediations” of speech in communication studies of language and the media. Once recorded, the human voice is generally seen as divested of its sociocultural codings (i.e., in traditional sociolinguistic terms of accent, dialect, etc.) when used in telephony and other industrial functions. Thus, for Weidman the forms of recorded voice we hear in media and computing put ideas of the authenticity of voice into question (Weidman, 2010, pp. 318-319). For Jeremijenko (Jeremijenko, 2005) recorded voices are subject to “social-technical systems” in which, as the result of technical processes of sound editing, they lose much of their human quality (such as hesitancy, pitch variation, and modulation in tone, “breathiness”). Jeremijenko argues that one of the key affordances of voice chip technologies is that they do not make any claims on us in terms of social identity:

Our exposure to voices (and other communicative sounds) that emanate from inanimate objects has become a significant part of our daily interactions: from radios to the more recent talking elevators, answering machine messages and pre-recorded music, television, automated phone menus, automatic teller machines, alarms and alerts, each of which . . . speaks in a language or dialect that makes little distinction between music, sound effects and articulated words, and privileges the situational function of language over the semantic and interactive. (Jeremijenko, 2005)

Recorded voices are here seen as radically detached from their social origins as they become articulated in the “techno-social landscape” (Jeremijenko, 2005). Indeed, one of the key rules in technician-designed IVR is of the need to “constrain” the impetus of callers to respond to “natural language” patterns or colloquial rhythms (Stentiford & Popay, 1999, p. 28) so that the system avoids stimulating expectations of normal conversation.

But even if IVR platforms technically manipulate and instrumentalize recorded voices to serve as “avatars” for company or organizational identities, it is still possible to examine how the process of technicizing of voice is itself subject to sociocultural determinations. Jeremijenko herself states that we cannot make a complete dichotomy between machine speech and social speech—that machine and human are coproducers of these systems (Jeremijenko, 2005). Similarly, Suchman (2004) argues that we need to think about the social indexing of “talking things”:

Voice chips, like other signs, rely very immediately upon and invoke their embedded locations for their intelligibility. Rather than things made in the image of the monadic, rational actor, we need reimagining of humans as contingently divisible participants in socio-material collectives who live out their particular histories in uniquely inflected ways. (pp. 264-265)

Similarly, Weidman notes how the soundscape of the media “creates new sonic encounters” (Weidman, 2010, p. 295) through which we might imagine social differences. Other writers note that the “residue of the social is part of the ‘uncanny’ feeling we experience from encountering recorded voices” (Neumark, Gibson, & Van Leeuwen, 2010). Writers concerned with the history of voice in the media such as Taylor or Hastings and Manning (Hastings & Manning, 2004; Taylor, 2009) also argue that we need to see the media and social forces as “coproducers” of voice. Hastings and Manning draw on Goffman’s ideas of “authors” and “animators” (2004, p. 309; Taylor, 2009) to underline the social roles present in producing speech in print and other media:

[Goffman] decomposes the speaker role into a production format of animator (the actual performer of the utterance), author (the one responsible for the text that it performed), principal (the one whose position is expressed), and figure. This final component—a kind of “embodied voice” which condenses a whole series of semiotic and social characterological features—is at once the least used and perhaps most useful for the analysis of the kind of realignments of voice, speaker and speech that we are interested in. (Hastings & Manning, 2004, p. 302)

The writers note the attribution or citing of “natural figure” voices in, for example, parody become animated by other agents and “shot through with alterity” (Hastings & Manning, 2004, p. 300). Indeed, most of the voices we hear are not in embodied form but are often staged or printed (i.e., in novels) and need to be evaluated in terms of mediated contexts.

Bakhtin, Dialogicism, and Voice

Seeing recorded voices as constructed, mediated but nevertheless inherently subject to sociolinguistic processes is taken up in this article by adopting Bakhtin’s ideas about “speech genres” (Bakhtin, 1986) and “dialogicism” (Bakhtin, 1984). Bakhtin notes how much of our spoken interaction is based on “pre-scription,” that even our mental concepts are linguistic in nature and our minds replete with others’ words (Bakhtin, 1986, p. 88). For Bakhtin all words are subject to “accentuation” (Morson & Emerson, 1990, p. 139), giving rise to a pervasive mimicry and echoing of others’ words, or inflexions of them in our speech (Bakhtin, 1984, p. 88). Bakhtin also sees relationality as fundamental to voice, that the utterance is “finalized” when it evokes a reply. However, intonation has the function of pointing the utterance beyond its immediate context (Bakhtin, 1986), endowing the utterance with social “addressivity” (Bakhtin, 1986).

Figure 1 from Morson and Emerson summarizes the broad outlines of Bakhtin’s dialogics and also has echoes of Goffman’s approach noted above. Direct discourse is often of the type of professional technical language articulated with a monological picture of audience in mind. Double-voiced words relate to direct narrative voicing, for example, putting words into a character’s mouth in a story. In “passive” voicing the “author/speaker” imitates someone else’s voice by “semantically inflecting” it (Morson & Emerson, 1990, p. 150) for a different purpose. Double voicing also occurs in parody, which is seen as an “arena of battle between two voices” (p. 193). But an original (“animating”) voice may sometimes spring back in the third type of active dialogics as it starts to undermine the parodic one and put it into a different light.

OPEN IN VIEWER

Figure 1.

Morson and Emerson (1990) on Bakhtin’s discourse types.

Monologic Voice: Social Indexing of Voice in Technician-Designed IVR

This section outlines the nature of the instrumentalizing of voice in technician-designed IVR services. Voice services technicians are telephonic engineers who deploy intuitive understandings of what they consider to be “appropriate” voices for the tasks or typical speech genres of IVR (such as “greeting,” “advising,” “directing,” “call finishing”). The IVR services telecoms industry is extensive and has been driven by companies wishing to avoid the high contract fees demanded by call-answering centers or to avoid employing large numbers of telephony staff. However, the setting-up of these services is quite expensive due to sound equipment, specialist microphones, and recording suites, and this means that most companies need to use voice services contractors to set up their IVRs. Although various dialogue standards exist in the technical side of the industry (i.e., “Dialogue 2000,” “Quickfuse”; Voxeo, 2012) and voice application editor software packages can be used to set general parameters (i.e., “Microsoft Speech App SDK”) for an IVR service, it remains technically demanding and usually voice technicians are the “authors” of the voices selected for these systems. Although the following quote from one independent voice contracting company seems to put the onus for selection on the client, it is clear that design and operation solutions are basically with the contractor:

Telecoms companies generally wait for the client to define the issue that needs resolving. Maybe at sales presentations they will go into the capacity of their particular system. But generally they will wait for the client to say, “We want three menu systems, one for in the office, one for when we are away from the office etc.” They will then come up with a solution to it—often, and then the client will not know about what else they could get. (Voice technician)

But in providing “solutions” technician-led voice services have to make a number of generally intuitive or experience-based choices about the selecting of voice, including the scripting of dialogue prompts/wording, and the type and range of “navigational aids” (Stentiford & Popay, 1999) required in the system.

Relational Weighting of Recorded Voices

In order to examine the choices of voices to be recorded and then associated with particular speech genres in IVR, six company IVR services were phoned up and recorded. Three of these were large national companies associated with home improvement and DIY, and three were large national bank/building societies. The “voice architecture” of each particular IVR platform was navigated as far as possible before one-to-one conversation with a human operator took over. This generally meant that there was a need to call the company quite a number of times depending on the complexity of the IVR menu structure so that variations in combinations of voices according to the range of caller queries could be explored.

When Bakhtin was writing in the 1950s he complained about the rudimentary nature of the available “taxonomies” of speech styles (Bakhtin, 1986, p. 64). However, there are now many elaborate taxonomies in “affective computing” (Nass & Brave, 2005; Scherer, Banziger, & Roesch, 2010), sociolinguistics (Crystal, 1975), prosodics (Bolinger, 1989), phenomenological (Ihde, 2007), and media studies (Haiman, 1985; Karpf, 2006).

The coding of voice in Table 1 below was adapted from the work of Abercrombie (1967) and Aronovitch (1976) and Eriksson (2007), who characterize voice on the basis of “voice cues” and characteristic “personality stereotypes” (Aronovitch, 1976, p. 208) or “conversational indexicals” in voice (Eriksson, 2007). Aronovitch gives a key role to pitch and loudness in categorizing voice (Aronovitch, 1976, p. 218) and refers to voice personalities such as “extrovert-introvert,” “kind-cruel,” and “bold-cautious.” Any perceived differences in degrees of pitch, tempo, and pause in IVR messages are derived from key prosodic indicators of voice in natural speech, those having a paralinguistic function of establishing “semantic stability between speakers’ meaning and intention” (Kennedy, 1991, pp. 8-9). Thus, although IVR is not naturally occurring speech its “prosodic units” or baseline variations in pitch, tempo, and pause still serve to engage the caller (Eriksson, 2007), and these recorded voices that suggest “personality stereotypes” (Aronovitch, 1976).

OPEN IN VIEWER

Table 1.

Summary of the Relational Weightings of Voice Indicators in IVR Answering Services.

Company & voice architecture	Age	Gender	Group/accent	Tempo/allegro	Hesitancy pause-breath	Tones -pitch	Speech genre
Homebase
Gateway	Mature	F	LE	Moderate	No	Normal	Welcoming and command
Subset	Young	F	Regional	Normal	Yes	Normal	Routine
Hold	Mature	F	LE	Slow	No	parenthetic	Therapy
Wilkinson
Gateway	Mature	F	LE	Moderate-fast	No	High	Command & routine
Subset	Young	F	LE	Medium-Fast	Low	Low	Routine
B&Q	Young	F	Regional	Regular	No	Low/soft	Command & routine
Santander	Mature	F	LE	Medium	Low	Medium	Advisory & command
HSBC	Mature	F	LE	Medium	No	Medium	Advisory & command
Nationwide	Mature	F	LE	medium	No	Medium	Advisory & command

Note: LE – London English.

Therefore, the values of “high,” “medium,” and “low” in Table 1 make no more claim than to be impressionistic, based on relational weightings of the various indices of voice. This means the focus was on the polarities perceived within the terms of the recorded IVR prompts alone, rather than exterior values based, for example, on voice spectrography or sociolinguistic phonetic markup. Recent ideas that voice belongs more to phonology than phonetics (Cavarero, 2005), that the significatory materiality of voice is “supra-segmental” and not accounted for in traditional linguistics but that it nevertheless has “performativity” (Dolar, 2006) support this idea of voice. Similarly, Bourdieu advises that we should avoid looking for distinctions in language that have little significance in light of the more sociologically significant fault lines occurring in the linguistic field (Bourdieu, 1991, p. 89). Following this approach avoids seeing intonation, for example, in naturalistic terms as an established variable rather than a constructed or relational one (in contrast, Crystal, for example, refuses to discuss anything about intonation that is not observable; Crystal, 1975, p. 30).

Main Contrasts in Voices Associated With Speech Genres in DIY and Banking IVR

Dialogic voice: Voice Branding Agencies

Voice branding agencies provide copy for “on-hold marketing” in which voice is deployed in much more consciously “creative” (Kotelly, 2003) ways. Voice branding is one of the key “intermediary” (Gress & Bagchi-Sen, 2007, p. 565) institutions in broader processes in the instrumentalizing of the human voice. Gress and Bagchi-Sen’s study of these agencies in South Korea notes that they act as “quasi-facilitators,” coordinating vast networks of “voice actors” involved in “local sound production industries” (p. 567). In terms of the significance of this industry, one of the companies interviewed in the research noted that they produce around 700 voice samples a month in response to inquiries from potential clients. The interviews with people in this industry show how the generally “monologic” nature of voices selected for the speech genres in technician-led IVR are often objects of ridicule. Voice branding stresses the need to “maintain caller interest” and to use the waiting time to inform customers of more than just basic information. But just as in technician-developed IVR, voice branding agencies also tend to rely on social stereotypes of voice.

Agents interviewed in these services were generally critical and dismissive of technician-led IVR services:

(In) audio branding we want to give clients” callers an incredible sense of the company in contrast to when the caller comes from the website to make the call and it’s Fred answering. (Audio branding script writer)

The derogatory reference to choosing in-house staff (“Fred”) is common in branding IVR, usually dismissed as the voice of “Sarah from Accounts.” In contrast, audio branding extols the virtue of only using “real voices,” heavily engineered voice being seen as robotic and unlikely to fit the brand image of most companies:

Does the woman who does the London Underground really exist? We deal only with real speech. (Voice branding agency manager)

This does not, however, mean using nonprofessional speech. Branding agencies offer a wide range of voices as part of their premium services:

To be honest one, maybe two, voice over artists could supply all of what we wanted if that’s what we wanted—but every client is different so we don’t use the same voice-over artists a hundred times. We try to vary it. So we might do male and female prompts alternating and we really try to follow this. So the voice over artists record it at home and send us the file and we can turn things over within hours, we do it regularly and we’ve so many voice over artists we never have to wait for one to be free. (Voice branding agency manager)

As well as being dismissive of in-house voices audio branders deride technicians’ resort to scripting platitudes such as “Your call is important to us.”

So voice branding agencies reflect more on the type, nature and character of a voice that might be suitable for any particular speech genre. They are thus more interested in the wider role of prosody and sound:

We also deal with literary voice. We think a lot about how we sound, whether being read or listened to. When we look at our clients’ websites we imagine how they would sound. (Sound studio production manager)

There is an appreciation of the production values of subtleties in tempo and pitch—that a conversational style should be at normal conversational level:

In a practical sense we don’t want anything too slow because the ideas in each prompt is seconds long and our clients don’t keep their callers on hold for hours on end so that they can listen to a very slow prompt. There has to be a balance between [being] audible and you’ve got to be able to understand it, and keeping it at a normal conversational level. I think that is the main guidelines—keep it at a conversational speed. We occasionally send things back because we think they are too slow, but sometimes it’s the voice over artists’ interpretation of the brief [that is wrong]. (Audio branding studio technician)

On most voice branding services’ websites there are usually to be found sound files giving an extensive range of “voice talents” often classified in prosodic and personality-type terms (i.e., “mellow Scots,” “sporty female”). Needless to say, these natural-sounding voices are rarely untutored or completely unengineered.

But voice branding IVR is promoted as dialogic and creatively scripted in this way. Rather than following technicians’ standardized formats for menus, voice branders “author” voices dialogically to create entertaining messages aimed at maintaining callers’ interest:

Our data is heavily cleansed. We really target and research companies before we call. Six copy writers will research the client and will write the samples or full scripts. Then a short brief to the voice artists with notes on intonation and emphasis perhaps on certain parts. And copy have real ideas of how it will sound—it’s written not to be read but spoken. (Production manager, North West England Sound Studio)

In Bakhtin’s terms, the key function of voice branding services is to overcome the fundamentally monoglossic nature of the experience of being “on-hold” and to turn the experience into a more dialogic one that might even be pleasurable for the caller (as well as being a market opportunity for the company):

You can still talk to people even though they are on hold. (Voice branding company director)

However, on-hold services face problems in creating the impression of dialogue on the caller. They must avoid the natural tendency of the caller to complete or “finalize” (in Bakhtin’s terms) the on-hold utterance. Thus, on-hold messages rarely have significant pauses or falling off of stress, which might create an expectation in the caller that they can reply. The pace of on-hold messages is generally conversational, but occasionally, micro pauses are used to create the impression of dialogue, as this interviewee notes:

We sometimes give short, very short, silences in which the caller can mentally comment—so the caller is trying to answer the questions. (Copywriter)

By mimicking conversational style, voice branding agencies thus avoid finalization of an utterance. For example, on one on-hold company website the voice of the well-known British and Hollywood actor Brian Blessed is employed to illustrate how an essentially monologic service can be “double voiced” in on-hold:

Is that it? [Incredulous] Really? Leaving me here on my own to listen to this . . . er . . . er nothing. Nothing? All this – silence. It reminds me of a girl I once knew—bored me to tears, [patronizing:] the poor dear. [sound effect—double bleep of a call waiting tone] Ah, there you are! I’ve forgotten about you! [music — “Green Sleeves”]. [censorious:] No No, No not that! [sound effect—needle scratching across a record]. [Conspiratorial:] You didn’t did you? Mmm—only a bloody fool would choose that! It’s not hard to keep me amused. [patience straining:] I’ll wait—even if you are a tight arse. JUST TALK TO ME!! [whistles to himself] It’s lonely over here. [Intimate:] I’ll tell you a secret—I’M just like everyone else and with a bit of entertainment your customers will hold the line. And 20% will buy whatever you are selling. Oh, [mocking tone:] you’ve got a website! I’ll look at it—beats scratching my arse. You know me, I’m your customer. You know it’s not dead time when I’m waiting. IT’S ALIVE!!! On hold marketing: EMBRACE [sound effect—the word echoes] YOUR CALLERS! (90 seconds duration)

At the start, Blessed’s voice suggests he is in a mellow, reflective mood, and he muses on the problem of being kept on hold; this is humorous because the actor is famous for his stentorian voice. But this example shows the type of dialogics characterizing voice branding IVR services. The intonation and other prosodic cues alongside the “conversational” style of the voice are consciously dialogicized by voice branders to create the impression of a number of voices.

Conclusion: Social Indexing of Voice in IVR

The above analysis has sought to show how both technician and voice branding IVR seek to overcome the conversational vagaries that arise from voice as a form of naturally occurring speech: Effectively callers’ utterances need to be subordinated to the IVR’ “speech plan.” A caller’s predisposition to “expressive intonation,” based on instinctive principles of naturally occurring language, is suppressed by the underlying monologic principles of both technician and branding agency IVR voice prompts. In this way, both types of IVR services seek to avoid “typical compositional and generic forms of finalization” (Bakhtin, 1986, pp. 76-77) that mark everyday dialogue. IVR menus thus prepare us as listeners, and position us, to mentally accept the subordination of our voices to those of IVR. Nevertheless, technician-directed IVR revealed primary speech generic structures such as advising, commanding, and placating therapy but also social referencing or “addressivity” of voice in accentual and gender/age codes. However, branding agency IVR was found to be more consciously dialogical and aesthetically stylized, distancing itself from clichéd copy by drawing on more literary and secondary speech genres.

Although Bakhtin’s idea of speech genres separates primary and secondary forms, his dialogics also suggests that poetic and everyday forms of voice are interrelated (Bakhtin, 1986, p. 65). Accordingly, this article’s adoption of his approach to understand the instrumentalization of recorded voice has shown how “authors” such as voice technicians and branders act to different degrees to dialogicize IVR. Bakhtin sees secondary literary speech genres as being more stylized whereas plainer styles mark “business documents, military commands, verbal signals in industry” (p. 63), a contrast also found in different approaches to the instrumentalization of voice in branding and technician IVR.

As Ahearn notes, Bakhtin’s dialogics tends to see all texts as “multivocal and egalitarian rather than univocal and authoritarian” and that “structure emerges through situated action which gives a sense of democratic play in language” (Ahearn, 2001, p. 128). Bakhtin allows us to recognize how different voices may articulate different speech genres in a dialogical and relational way—that “in essence, meaning belongs to a word in the position between speakers” (Maybin, 2003, p. 69). But it is by adding Bourdieu’s approach to Bakhtin’s that a more critical assessment of the relation of particular speech genres with socially coded voices can be made. The bringing together of Bourdieu and Bakhtin’s ideas about language has largely been overlooked (but see Hanks, 1987) although both share a primary focus on language at the level of parole. And while there is no clear formulation of the idea of field in Bakhtin, he did see speech genres as having an existence “beyond context” (1986, pp. 85-86) —it was just in particular articulations that they become “accentuated.” Uniting these two approaches, then, has some epistemological validity and, in practice, has allowed this article to critically engage with the process of instrumentalizing voice in the contemporary voice industry.