Speech communication in real-world settings typically involves several sources of adverse conditions Speaker Environment Listener Bi-directional talker-listener Source Environmental / Receiver limitations: adaptation across a language barrier degradation: transmission degradation: • Peripheral deficiency • Conversational speech • Incomplete language model • Accented speech • Primarily energetic Ann Bradlow • Impaired language model • Disordered speech masking (e.g. broadband access /use Department of Linguistics noise) • Energetic & informational • Cognitive load Northwestern University masking (e.g. background speech) Speech communication across a language barrier (Mattys, Davis, Bradlow and Scott, Language & Cognitive Processing , SI on Speech Recognition in Adverse conditions, 2012.) Speech communication in real-world settings typically Why is foreign-accented speech hard to understand? involves several sources of adverse conditions Deviation of the signal from the native talker norm/target Speaker Environment Listener The children dropped the bag. Source Environmental / Receiver limitations: Native- degradation: transmission accented degradation: • Conversational speech • Peripheral deficiency • Incomplete language • Primarily energetic • Accented speech masking (e.g. broadband model • Disordered speech noise) • Impaired language Chinese- • Energetic & informational model access /use accented masking (e.g. background • Cognitive load speech) Speech communication across a language barrier • a challenge The foreign-accented sentence: • an opportunity for innovation ‣ is ~30% longer overall (lots of pauses, less fluent) ‣ exhibits different segmental/sub-segmental timing relations ‣ etc. (Mattys, Davis, Bradlow and Scott, Language & Cognitive Processing , SI on Speech Recognition in Adverse conditions, 2012.) 1
Systematicity of foreign-accented speech ”deviations„ Systematicity of foreign-accented speech ”deviations„ (2) L2 typological peculiarities => Accent-independent adaptation (1) L1-L2 interactions => Talker-independent adaptation Adaptation to foreign-accented speech by talkers from a Adaptation to an accent as it extends across a group of variety of native language backgrounds foreign-accented talkers from the same native language background Slovak Slovak English English Mandarin Mandarin Adaptation to foreign-accented speech L1-L2 interactions => Talker-independent adaptation Adaptation to an accent as it extends across a group of foreign- Study 1: Adaptation to systematic deviations of foreign-accented accented talkers from the same native language background speech following exposure to stimuli that vary along the to-be- learned dimension Test Training 1 Training 2 • Talker-independent adaptation Adaptation to an accent as it extends across a group of foreign-accented talkers from the same native language background • Accent-independent adaptation A. Chinese-accented test talker Adaptation to foreign-accented speech by talkers from a variety of native B. Multiple Chinese-accented talkers language backgrounds C. Single Chinese-accented talker Chinese-accented Study 2 : Adaptation to foreign-accented speech in response to D. Multiple native-accented talkers (in white noise, +5 dB SNR) variation in the training task Slovakian-accented E. Untrained controls • Does perceptual learning for foreign-accented speech require (in white noise, +5 dB SNR) active performance of a sentence recognition task? Bradlow and Bent, 2008 . See also Clarke & Garrett, 2004; Sidaras, Alexander & Nygaard, 2009. 2
L2 typological peculiarities => Accent-independent adaptation Talker-independent adaptation to a foreign-accent Adaptation to foreign-accented speech by talkers from a variety of native language backgrounds Test Training 1 Training 2 A. Chinese-accented test talker B. Multiple Chinese-accented talkers C. Single Chinese-accented talker Chinese-accented D. Multiple native-accented talkers (in white noise, +5 dB SNR) E. Multiple accents (Chinese, Romanian, Thai, Hindi, Korean) Slovakian-accented (in white noise, +5 dB SNR) F. Untrained controls Adaptation to foreign-accented speech Accent-independent adaptation to a foreign accent Post test 1: Chinese-accented talker Multiple accent training : 100 Study 1: Adaptation to systematic deviations of foreign-accented ‣ Chinese ( � test talker) 90 speech following exposure to stimuli that vary along the to-be- ‣ Romanian Percent Correct learned dimension ‣ Thai 80 ‣ Hindi 70 • Talker-independent adaptation ‣ Korean 60 Adaptation to an accent as it extends across a group of foreign-accented talkers from the same native language background 50 • Accent-independent adaptation Multi- Multi- Test talker Single Native Untrained Accent talker talker talker Adaptation to foreign-accented speech by talkers from a variety of native 100 Post test 2: Slovakian-accented talker language backgrounds 90 Percent Correct Study 2 : Adaptation to foreign-accented speech in response to 80 variation in the training task 70 60 • Does perceptual learning for foreign-accented speech require active performance of a sentence recognition task? 50 Multi- Multi- Test talker Single Native Untrained Accent talker talker talker Baese-Berk, Bradlow & Wright, 2013. 3
Auditory perceptual learning with a combination of Auditory perceptual learning with a combination of active task performance and passive stimulus exposure active task performance and passive stimulus exposure worse worse � Pretest � Pretest 16 16 Frequency-discrimination Frequency-discrimination � Post-test � Post-test threshold (Hz) threshold (Hz) 10 10 6 6 better better Frequency Training Frequency Training Frequency Training Control Control Frequency Training + (untrained) (untrained) + Stimulus Exposure Stimulus Exposure Wright, B.A., Sabin, A.T., Zhang, Y., Marrone, N., & Fitzgerald, M.B. (2010), J. Neuroscience. Wright, B.A., Sabin, A.T., Zhang, Y., Marrone, N., & Fitzgerald, M.B. (2010), J. Neuroscience. Adaptation to foreign-accented speech with a combination of Post-test 1: Chinese-accented Talker Training: Multi-talker 100 active task performance and passive stimulus exposure 90 Test: Mandarin-accented Talker 80 70 60 50 Multi-Accent Multi-talker Test talker Single talker Native talker Untrained All Active Active Training Active+passive Passive Exposure All Passive Passive Task (Silence) Short Active Post-Test “Passive” task: Active+Passive training results in as much learning as All-Active training. 4
Adaptation to foreign-accented speech Talker-listener interaction: Spontaneous conversational patterns across a language barrier • Systematic deviations of foreign-accented speech allow highly The Diapix task (dialogue-based picture matching) generalized perceptual learning with exposure to appropriately ‣ A ”spot-the-difference„ game with 2 pictures and 2 participants. variable training stimuli. ‣ Without seeing each other‚s picture, participants work together to find differences. ‣ Elicits a wide range of utterance types (questions, declaratives, exclamations etc.). • Adaptation to foreign accents can occur in response to a ‣ Elicits connected speech from both participants without predetermined roles. combination of active performance of a sentence recognition task and passive listening situations. Picture B Picture A • Perceptual flexibility underlying perceptual adaptation to foreign- accented speech may eventually lead to parallel adaptations in speech production. => A link between individual-level adaptation to variable speech input and population-level, contact-induced sound change. Communicative efficiency, phonetic convergence, Communicative efficiency and language distance and language distance Task accuracy Time to complete the diapix task Communicative efficiency Experimenter imposed time limit • Task completion time • Type-to-token ratio Phonetic convergence • Talker similarity judgments at the beginning versus at the end of a conversation Language distance Close Far Close Far NN 1 -NN 1 N 1 -N 2 N-NN NN 1 -NN 2 N 1 -N 1 N 1 -N 2 N-NN NN 1 -NN 1 NN 1 -NN 2 N 1 -N 1 Efficiency decreases with increasing language distance. Van Engen, Baese-Berk, Baker, Choi, Kim & Bradlow, , 2010. See also Baker & Hazan, 2011; Hazan and Baker, 2011. 5
Recommend
More recommend
