Infant Vocal Imitation of Music

Participant

James, a 15-month-old male infant, lived with his mother and father, 5-year-old brother, and 3-year-old sister in a middle-class urban neighborhood of a large city in the southwest region of the United States. The mother worked as a homemaker and the father as a software engineer. Both parents held graduate degrees and had studied music. The father had participated in band between 4th and 12th grades and had played saxophone and clarinet in a garage band while in high school. The mother had taken piano lessons as a child for 10 years. All three children participated in music classes for young children: James and the sister attended a Suzuki music class for children ages 0 to 3 and their parents, and the brother participated in a string program. It was through these classes that the family was recruited to participate in the study. We did not select James and his family for a particular motive; although it was clear that the parents were interested in having their children learn music, we had no reason to believe that James’s behavior would be in any way special. For recruitment purposes, the parents were told that the study focused on infants’ home sound environment. They were aware that the main areas of interest were music and language, but they were not told that the researchers were focusing specifically on infant singing. The parents provided consent to participate following institutional review board guidelines.

Data Collection

The infant wore a small and light recording device throughout an entire day. The recording device, known as Digital Language Processor (DLP), is part of a system developed by LENA (Language Environment Analysis; www.lena.org) for studying the home language environment of infants and young children. The DLP captures sounds produced within hearing range of the infant, including the infant’s own vocal production, for up to 16 continuous hours. The audio data are analyzed by the LENA software, which provides estimates of language indicators, such as total number of adult words, total number of conversational turns, child vocalization frequency, and total time of ambient sounds (e.g., sounds from electronic devices, television, and toys). Estimates of music data, such as ambient music, singing by adults, or infant vocal production of music, are not provided; it is necessary to listen to the audio file to identify and study such music indicators. Although originally developed for language research, the use of LENA technology is also valuable for gaining insight into infants’ and young children’s early musical experiences (Costa-Giomi, 2016; Costa-Giomi & Benetti, 2017; Costa-Giomi & Sun, 2016; Dean, 2015; Fausey & Mendoza, 2018).

In the present study, we investigated James’s music environment and imitative vocalizations by listening to the complete audio file of the infant recorded on a weekday during which the family stayed at home. The mother turned on the device when James woke up in the morning and turned it off 16 hr later. The parents provided a time log listing the main activities of that day, such as nap times and meals (see Table S3 included with the online version of the article).

Data Analysis

We first listened to the audio file and identified the events reported by the parents (see Table S3 in the online version of the article). We then divided the audio file into shorter segments to make the listening of the 16-hr file more manageable. Each segment corresponded loosely to an event or a general interaction that occurred on that day (e.g., mom and children playing in the living room, dad and infant walking outside). Each segment was subjected to repeated listening, first to describe the general context, such as who was present and where they were in relation to the infant, and then to register in greater detail the characteristics of interactions, vocalizations, and any music heard by the infant. A description of each segment is provided in Table S3 in the online version of the article. From these segments, we isolated smaller sections that contained music and/or infant vocalizations for in-depth listening. Infant vocalizations and music directed at James or in the background were described and transcribed using music symbols when appropriate. We also identified the repertoire James was exposed to (e.g., artist and song names) and the source of the music (e.g., sister singing, electronic toy, brother plucking cello). The entire procedure involved several months of repeated listening, segmentation of the audio file, taking notes, registering the characteristics of the music environment, and isolating sections for analysis.

It was through this systematic process that we identified the two song imitations that are the focus of the present study. To clarify, we defined song imitation as a vocalization that was perceptually similar to a song model present in the infant’s sound environment. To corroborate our pairing of the infant’s imitations and the song models he had heard, we examined whether listeners who were not involved in the study and were unaware of the source of the data and study objectives also would perceive the similarity between the infant vocalizations and the song models. The responses of the 65 adults who participated supported our identification of the imitations and song models. A full description of the questions participants answered, the stimuli they heard, and their answers is provided in the supplemental material (included with the online version of the article).

Imitation 1: “Rain Rain”

Shortly after lunch, James heard his mother sing “Rain Rain” twice. The music notation is shown in Figure 1a, with the fundamental frequency (F0) values of the two notes that compose the initial minor-third interval indicated below the notation. The first execution was in the key of A-flat major and the second execution was in the key of A major.

OPEN IN VIEWER

Figure 1.

“Rain Rain” imitation. (a) Music notation representing the mother’s singing, (b) the infant’s first vocalization, (c) the father’s singing and infant imitation of the father, and (d) the father’s singing during the last part of the infant–father interaction. The pitch contours of individual infant sounds are shown below the musical notation in (b).

Six hours later, in the evening, when James was alone with his father, he vocalized a sequence of 11 discrete sounds that resembled the rhythm and pitches of “Rain Rain.” The sequence lasted approximately 7 s and comprised syllabic combinations of consonant-like sounds and vowels. The pitches of the first two sounds were E-flat4 and C4, respectively. The onsets of the sounds occurred at regular intervals except for the last three sounds, when James seemed to have some difficulty coordinating his breathing. The durations between consecutive onsets clustered around two categories of values in a ratio of 2:1, producing a temporal organization of sounds similar to that of the rhythm of the song. The predictability of timing induced the sensation of an isochronous pulse of approximately 78 beats per minute. James’s breaths occurred at the end of every second beat and naturally divided the sequence into four groups of approximately equal duration. The first sound in each group was the loudest, giving the impression of a metric accent at the beginning of the group.

The rhythmic pattern and the initial pitches of James’s sequence of sounds were perceptually similar to the rhythm and melody of the song “Rain Rain” that his mother sang earlier that day. Figure 1b depicts the infant’s sequence of sounds using music notation in the key of A-flat major, with the four groups represented as four measures and the duration categories as half notes and quarter notes. F0 values are indicated below the music notation, and the duration of each sound and each group are indicated above the notes and measures, respectively. In each of the four groups, the descending melodic contour was preserved. The pitch contours of each sound are shown below the music notation in Figure 1b. As can be seen in this figure, the pitch contour of each sound was either mostly stable (e.g., the contour of the first sound) or centered on a pitch target (e.g., the slight rise and fall of the contour of the third sound). The durations of the inter-onset intervals (i.e., time elapsed between the onsets of successive sounds) clustered around two values, 0.4 s and 0.8 s, which is consistent with the perception of two duration categories. The duration of each of the four groups was approximately equal, ranging from 1.63 s to 1.80 s (M = 1.69 s), which supports the perception of the vocalization as comprising four approximately equally timed groups.

The father, who was next to James when he vocalized, identified the resemblance between the infant’s sounds and the song “Rain Rain”: Immediately after James finished the vocalization of interest, the father said, “It sounds like you’re singing . . .,” and proceeded to sing the song himself, in the key of G major and starting on D3 and descending to B2. Figure 1c shows the music notation for the father’s singing, with F0 values for the first two notes indicated below. James responded with a sequence of vowels that matched the rhythm and melodic contour of the last measure of the song sung by the father (Measure 4 of Figure 1c). Figure 1c also shows the infant’s vocalization represented with music notation and the corresponding F0 values. The first three sounds formed a descending interval pattern with durations equivalent to two quarter notes followed by a half note. James did not begin on E-flat4, as he had done in his initial vocalization, but rather on D4, an octave higher than his father’s starting pitch. The lower note of the interval was not clearly pitched, but it was perceptibly lower than D4. Despite the somewhat unclear pitch percept induced by the infant’s lower note, inspection of the waveform showed a periodic pattern for this sound, with F0 of 233 Hz (B-flat3). Thus, even though James was not entirely successful in matching the pitch class of the father’s lower note (B) as he was with the higher note of the minor third interval (D), the interval he produced was quite close to the father’s, and the descending melodic contour of the tune was preserved.

James’s following two sounds maintained the rhythm of “Rain Rain,” but the sounds were partially obscured as the father interrupted and began singing again. This time, however, his father sang the melody a half step lower, in the key of G-flat major, starting on D-flat3 and descending to B-flat2 (Figure 1c; “Come a-gain a-no-ther day”). When the father finished, James produced three syllable sounds that matched the rhythm of the last part of his father’s singing (i.e., “-no-ther day”) but with unclear pitch (Figure 1c). The three syllables were perceptually similar to the sounds “no,” “der,” and “deh,” matching parts of the accompanying lyrics. James followed this with another two sounds that resembled “da da.” When James stopped vocalizing, his father resumed singing the same melody, maintaining the key of G-flat major but with variations in lyrics (Figure 1d). After his father’s fourth execution of the melody, James vocalized three sounds that resembled the rhythm of the last three notes sung by his father, followed by short moans of frustration and crying. The father sang the melody one last time, again with alternate lyrics. After a few seconds of silence, the infant changed the focus of the interaction by making “uh-oh” sounds, and the musical exchange between father and infant ended.

In summary, the results show that the infant produced an unprompted vocalization that resembled the song his mother had sung 6 hr earlier in interval structure, key, rhythm, and temporal organization (i.e., meter). He then engaged in an imitative interaction with his father that centered on the same song and reflected speech and musical elements, such as consonants and vowels from his father’s lyrics, interval structure, key, and rhythm.

Imitation 2: “Happy Birthday”

In the morning, James interacted with an electronic music toy that played the melody “Happy Birthday.” During the 10-min interaction, the toy played 11 executions of the entire melody and one partial execution, for a total of 128 s. The melody was in G major, starting on D4, and consisted of synthesized keyboard sounds and no lyrics (Figure 2a). The tempo was approximately 140 beats per minute (duration of quarter note = 0.43 s). In the evening, while with his parents, James vocalized 10 sounds that resembled the beginning of “Happy Birthday.” The entire vocalization lasted slightly over 4 s. The parents were talking among themselves while the infant vocalized.

OPEN IN VIEWER

Figure 2.

“Happy Birthday” imitation. Music notation representing the (a) toy melody and (b) infant’s sounds.

The music notation of the infant’s sounds is shown below the toy melody in Figure 2b, and F0 values are shown below the notation. The infant’s first seven notes matched the pitches of the melody played by the toy. Of these seven notes, only G4 seemed to present some difficulty: After the jump from D4, James overshot but then corrected the note by gliding down a half step. Figure 2b also shows the durations of each of the infant’s notes. The temporal relationship between the note durations is represented using quarter notes in the musical notation. The tempo was approximately 130 beats per minute (average duration of quarter note = 0.46 s).

James produced the first four notes (D4, D4, E4, and D4) with different vowel sounds. They were perceptually similar to the vowels of the words “bad” (D4), “bed” (D4), “bud” (E4), and “bead” (D4) in American English. Acoustic analysis showed changes in formant patterns consistent with the perception of different vowel qualities (see the supplemental material online for more details and results of acoustic analysis). The subsequent notes were produced with sounds similar to “ya,” except for the seventh note (D4), which was produced with a “ga” syllable.

In conclusion, perceptual and acoustic analyses of the infant’s sounds indicate that James reproduced the pitches of the beginning of the melody of the song “Happy Birthday” played by an electronic toy and incorporated different vowel sounds in his vocalization.

“Rain Rain”

James vocalized a sequence of sounds that resembled the melody of “Rain Rain,” a song that his mother had sung 6 hr earlier. James’s unprompted vocalization displayed an organized temporal pattern and prominent descending minor-third characteristic of the song. The father heard James sing the melody and proceeded to sing the song himself; James and his father then engaged in an imitative interaction with the father singing various verses of the song and James imitating intervals and rhythms. There are several points worth emphasizing about James’s vocalizations, the parent’s vocalizations, and the interaction in general. Concerning the infant’s vocalizations, (a) James’s initial vocalization expressed features relevant to musical organization, such as discrete sounds with stable pitch contours and with durations conforming to music rhythm. That the father labeled the song so promptly after hearing the infant’s vocalization suggests that James was able to express critical musical elements of the song, facilitating his father’s identification of the melody. (b) The infant’s second “Rain Rain” vocalization—the one following the father’s singing of the song—reflected the same descending melodic interval of the first vocalization but was sung in the key of G major instead of the original A-flat major. James transposed the melodic interval to match his father’s rendition of the song. We know from perceptual studies that infants can encode both absolute and relative pitch information from music melodies (Plantinga & Trainor, 2005; Saffran, 2003; Saffran et al., 2005; Saffran & Griepentrog, 2001; Trainor & Trehub, 1992) and that the learning and encoding of pitch depends on the information available to them (Saffran et al., 2005); here, we describe how one infant’s vocalizations reflected the transposed melodies he heard around him.

Second, regarding the parents’ singing, it is worth noting that the mother and father provided James with models of the same melody sung in four different keys (A major, A-flat major, G major, and G-flat major), by two different voices, and with different tempi—all on the same day. That James had multiple opportunities to hear the same song sung in different ways is important from a music development perspective because such variability supports the development of melodic categorization (Costa-Giomi, 2013). For example, by being exposed to the same song sung in various keys or by different voices, James had the opportunity to perceive the similarities of the melody despite differences in key or timbre. This type of input prioritizes the structural pitch information of the melody (Patel, 2008) because such information is what remains unchanged amid variations in pitch or timbre. Adding to further sources of input variation, the parents’ singing was not professional; that is, although the song was communicated effectively, there were moments in which notes were out of tune, the pitch drifted, and the timings were slightly off. Considering that the identity of a melody should be robust to all these eventual variations, it may be the case that input representative of natural singing with all its variability is particularly beneficial for music learning as it, too, reinforces the formation of music categories.

Finally, the “Rain Rain” episode is an illustration of a musical interaction common between infants and caregivers that emphasizes communication and emotional connection (Costa-Giomi & Benetti, 2017; Dissanayake, 2009; Malloch, 1999). James initiated the musical exchange by calling his father and vocalizing; the father acknowledged and elaborated on the infant’s music vocalizations, in effect affirming the infant’s vocalization as a successful and valid expression of music. The song then served as a framework for coordinated turn taking between infant and father, much like the communicative musicality of mother−infant dyads described in previous research (Malloch, 1999). The continuous recording of James’s soundscape revealed that the same song was part of two meaningful infant−parent events during that day, one with the mother after lunch and one with the father in the evening. In other words, the song was not an isolated occurrence and was not bound to one single interaction but was a musical object shared by at least three family members and embedded into James’s daily life. The results of the analysis of “Rain Rain” highlight the value of the particular methodology we used. Not only did it allow us to establish that James had heard the song sung by the mother earlier—information that, arguably, we could have obtained from a report by a very attentive parent with excellent memory—but it also let us determine the exact duration, key, and tempo of both the adults’ singing and the infant’s singing.

“Happy Birthday”

The second infant vocalization of interest consisted of a melody sung by James that matched the pitches of “Happy Birthday.” James had heard this melody produced by an electronic toy with which he had played 10 hr earlier. Although James heard the melody played by the toy with a synthesized timbre and no lyrics, he incorporated vowel sounds in the first four notes of his vocalization. Acoustic analysis showed changes in formant patterns consistent with the perception of different vowel qualities. Considering that James did not hear the melody sung with words on that day, the finding that he produced sounds resembling linguistic content in addition to musical content is notable. It is possible that the vowels he produced were simply the result of vocal exploration, but this argument does not detract from the fact that he produced perceptually and acoustically distinct vowel sounds as he imitated the pitches played by the toy. The possibility that he associated the melody from the toy with sung renditions of “Happy Birthday” cannot be dismissed. After all, “Happy Birthday” is a popular song, and it is likely that James heard it sung many times before the day of data collection.

Unlike live singing, toys provide many opportunities for repeated listening. At the kick of a crib mobile or the push of a button, infants can hear melodies at will as many times as they wish (Costa-Giomi & Merkow, 2015; Young 2008). In fact, James was engaged in playing with the toy for 10 min and heard the melody that he later imitated played 12 times in quick succession. Considering the abundance of interactive music toys and devices in the homes of young children (Young, 2008), it seems important to study their impact on infant music learning and singing development. Studies of infants’ engagement with digital music toys suggest that adult–infant joint play with toys may provide meaningful musical experiences (e.g., Costa-Giomi & Merkow, 2015; Merkow, 2013; Young, 2008), but we know little about how infants’ independent play with such devices affects music learning and, more specifically, singing. The methodology we used in the present study allowed us to capture a moment of independent play and the vocalization that occurred hours later. Without this extended and continuous recording, we would have missed James’s rendition of the song, just like the parents did. Being able to study infants’ spontaneous singing as it occurs when they are not being observed is one of the advantages of using devices such as LENA.

Music Learning

Although much is known about perception of music by infants, little is known about how such perception skills interact and translate into observable manifestations of musical understanding. Arguably, when producing music, perceptual capabilities become crystalized: Models of accurate singing in adults propose that music imitation involves, among other things, perceiving the relevant features of the music model and organizing an appropriate motor response to match them (e.g., Berkowska & Dalla Bella, 2009; Pfordresher et al., 2015). We can speculate that James’s organized expression of identifiable music features was in part a reflection of his music perception abilities. He produced clearly articulated and resonant sounds with stable pitches, and his vocalizations showed temporal organization and a regular pulse. James’s imitations were not flawless renditions of the songs, but they were similar enough to the models that adult listeners, such as ourselves, the participants in the perception study, and James’s father, were able to perceive the similarity. The very fact that his vocal imitations were not at par with adult singing standards calls for a framework of early singing that contemplates a developmental trajectory of music vocalization that takes into consideration the developmental status and limited experience with music and music making of infants. ¹

The technology presented here may facilitate the systematic study of such vocalizations as they emerge during the first 2 years of life. It allows for studying manifestations of infant vocalization as they occur in their natural environments, including vocalizations that occur during moments of independent play or that may go unnoticed by parents. The technology also makes it possible to integrate the study of the music environment and the development of vocalizations. The connection between what infants hear and what they learn has been studied in the context of language but not in the context of music. For example, we know that infants learn to produce words that are part of their language environment and that their everyday language experiences are associated with specific language outcomes (Roy et al., 2015; Swingley & Humphrey, 2018). The analysis of all the sounds produced by and around the infant continuously may facilitate the study of these types of connections for the understanding of singing development.

The results of the present study allowed us to trace connections between music environment opportunities and specific singing outcomes in one infant. We recognize that it is possible that James’s music vocal behavior was exceptional and not representative of infant singing in general. His music environment was indeed particularly rich: He heard multiple instances of infant-directed models of songs; engaged in imitative, turn-taking singing with an adult (who acknowledged and celebrated his music vocalization); and had opportunities to produce music by himself without adult intervention. Would infants raised in homes with fewer opportunities for music engagement show comparable vocalization behaviors? Exploring the connections between the music environment and the vocalization of infants will provide answers to this question. Such research will not only inform us about how to shape infants’ music environment to optimize learning opportunities and support singing development but will also provide insight into the mechanisms and developmental processes underlying infants’ ability to learn from everyday music experiences.