Continuous vocal imitation with self-organized vowel spaces in Recurrent Neural Network

Kanda, Hisashi; Ogata, Tetsuya; Takahashi, Tôru; Komatani, Kazunori; Okuno, Hiroshi G.

doi:10.1109/robot.2009.5152818

Cited by 14 publications

(10 citation statements)

References 16 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Kanda et al [3] proposed a continuous vocal imitation system based on a recurrent neural network with parametric bias (RNNPB) that explains how infants acquire phones.…”

Section: A Non-interactive or Interactive Cases With Homogeneous Agentsmentioning

confidence: 99%

“…An animatronic model of the human tongue and vocal tract, called "AnTon," was designed by Hofe and Moore [66]. They reproduced human speech gestures based on AnTon's tongue control 3 . Because their main purpose was to investigate animatronic control, the quality of the reproduced sounds was not discussed.…”

Section: E Research Platformsmentioning

confidence: 99%

“…Some of this modeling has not explicitly addressed the issue of interaction between an infant and its caregiver (e.g., [2], [3], [4], [5], [6], [7], [8], [9], [10]). Others have attempted to explain an infant's imitation ability based on acoustic matching (e.g., [11], [12]).…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Modeling Early Vocal Development Through Infant–Caregiver Interaction: A Review

Asada

2016

IEEE Trans. Cogn. Dev. Syst.

View full text Add to dashboard Cite

The developmental origin of language communication seems to involve vocal interactions between an infant and a caregiver, and one of the big mysteries related to this is how an infant learns to vocalize the caregiver's native language. Many theories attempt to explain this ability of infant as imitation based on acoustic matching. However, the acoustic qualities of speech produced by the infant and caregiver are quite different and therefore cannot be fully explained by imitation. Instead, the interaction itself may have an important role to play, but the mechanism is still unclear. In this article, we review studies addressing this topic based on explicit interaction mechanisms using computer simulations and/or real vocal robots. The relationships between these approaches are analyzed after a brief review of the early development of an infant's speech perception and articulation based on observational studies in developmental psychology and a few neuroscientific imaging studies. Finally, future issues related to real infant-caregiver vocal interaction are outlined.

show abstract

“…Kanda et al [3] proposed a continuous vocal imitation system based on a recurrent neural network with parametric bias (RNNPB) that explains how infants acquire phones.…”

Section: A Non-interactive or Interactive Cases With Homogeneous Agentsmentioning

confidence: 99%

Section: E Research Platformsmentioning

confidence: 99%

See 1 more Smart Citation

Modeling Early Vocal Development Through Infant–Caregiver Interaction: A Review

Asada

2016

IEEE Trans. Cogn. Dev. Syst.

View full text Add to dashboard Cite

show abstract

“…Kanda et al (2009) have also addressed learning to produce the vowels of a given language. The model is a recurrent neural network with parametric bias (RNNPB).…”

Section: Introductionmentioning

confidence: 99%

Prespeech motor learning in a neural network using reinforcement

et al. 2013

View full text Add to dashboard Cite

Vocal motor development in infancy provides a crucial foundation for language development. Some significant early accomplishments include learning to control the process of phonation (the production of sound at the larynx) and learning to produce the sounds of one’s language. Previous work has shown that social reinforcement shapes the kinds of vocalizations infants produce. We present a neural network model that provides an account of how vocal learning may be guided by reinforcement. The model consists of a self-organizing map that outputs to muscles of a realistic vocalization synthesizer. Vocalizations are spontaneously produced by the network. If a vocalization meets certain acoustic criteria, it is reinforced, and the weights are updated to make similar muscle activations increasingly likely to recur. We ran simulations of the model under various reinforcement criteria and tested the types of vocalizations it produced after learning in the differ-ent conditions. When reinforcement was contingent on the production of phonated (i.e. voiced) sounds, the network’s post learning productions were almost always phonated, whereas when reinforcement was not contingent on phonation, the network’s post-learning productions were almost always not phonated. When reinforcement was contingent on both phonation and proximity to English vowels as opposed to Korean vowels, the model’s post-learning productions were more likely to resemble the English vowels and vice versa.

show abstract

“…After this random exploration, it is given specific phonetic targets and further fine tunes its synaptic connections, learning to produce new speech sounds and sequences of speech sounds. Relatedly, learning to produce particular sequences of vowel sounds based on experienced input sequences has been addressed in work by Kanda et al [16].…”

Section: B Existing Computational Models Of Infant Vocal Developmentmentioning

confidence: 99%

A spiking neural network model of canonical babbling development

Warlaumont

2012

2012 IEEE International Conference on Development and Learning and Epigenetic Robotics (ICDL)

View full text Add to dashboard Cite

The author wished to add the following correction:"After publication, it was discovered that the model did not receive reinforcement in the way that the paper described. The main conclusions still hold after correcting the error and making the following changes: performing synaptic weight normalization in place of spike-timing-dependent depression, removing the constant low level of dopamine, and scaling muscle activity by 2.5. A simulation and yoked control of the corrected model exhibited an increase in performance over time in the main simulation (r = 0.096, p < 0.001) as well as, to a lesser extent, in the control (r = 0.070, p = 0.008), and performance was significantly higher overall in the main simulation compared to control (T = 1.870, p = 0.031)."Abstract-Canonical babbling, the production of vocalizations that contain mature-sounding syllables, is one of the most striking and important milestones prior to the production of first words. This study simulates the emergence of canonical babbling using a spiking neural network containing motor neurons that activate muscles in a vocal tract simulator. The spiking neural network periodically produces synthesized vocalizations and a human listener judges the vocalizations on the basis of their syllabicity, deciding whether or not to reward the model. If a reward is given, spike timing dependent plasticity is increased and the model becomes more likely to recreate a pattern of neural firings similar to that which generated the reinforced vocalization. The model successfully increases its production of mature-sounding canonical syllables, whereas a yoked control simulation does not exhibit any such effect. This finding corresponds to results of experimental work with human infants in which consonantvowel syllable production is selectively reinforced by the infants' caregivers.

show abstract

Continuous vocal imitation with self-organized vowel spaces in Recurrent Neural Network

Cited by 14 publications

References 16 publications

Modeling Early Vocal Development Through Infant–Caregiver Interaction: A Review

Modeling Early Vocal Development Through Infant–Caregiver Interaction: A Review

Prespeech motor learning in a neural network using reinforcement

A spiking neural network model of canonical babbling development

Contact Info

Product

Resources

About