Figure 1 - uploaded by Jaekoo Kang
Content may be subject to copyright.
The confusion matrices for 9 vowel categories from the forward mapping model. Accuracies are annotated in each cell. Top: result from a female native speaker (JW14). Bottom: result from a female L2 speaker (F36).
Source publication
Variability is widespread in speech, but it is unlikely that all of it is harmful; variability in other domains has been shown to allow flexibility, within limits. Using one technique for separating the two, we applied Uncontrolled Manifold Analysis (UCM) to vowels in running speech. This results in two multidimensional manifolds, one the controlle...
Contexts in source publication
Context 1
... mean accuracy of our neural-net forward mapping models for the native speakers was 78% (SD=5%), tested on the held-off test set (20% of the total data). Figure 1 demonstrates the confusion matrices for a female native speaker (JW14) and a female L2 speaker (F36), both of whom were in their mid-thirties. As expected, the mean accuracy for F36 (L2) (39.18%) is lower than that for JW14 (L1) (85.47%). ...
Context 2
... expected, the mean accuracy for F36 (L2) (39.18%) is lower than that for JW14 (L1) (85.47%). In particular, the English vowels pairs /i, ɪ/ and /ɛ, ae/ each map on to a single vowel in Korean [19] and were more susceptible to be confused than others for Korean speakers as shown in Figure 1 (58% for /i/, 33% for /ɪ/, 16% for /ɛ/ and 37% for /ae/). The confusion matrices for 9 vowel categories from the forward mapping model. ...
Context 3
... mean accuracy of our neural-net forward mapping models for the native speakers was 78% (SD=5%), tested on the held-off test set (20% of the total data). Figure 1 demonstrates the confusion matrices for a female native speaker (JW14) and a female L2 speaker (F36), both of whom were in their mid-thirties. As expected, the mean accuracy for F36 (L2) (39.18%) is lower than that for JW14 (L1) (85.47%). ...
Context 4
... expected, the mean accuracy for F36 (L2) (39.18%) is lower than that for JW14 (L1) (85.47%). In particular, the English vowels pairs /i, ɪ/ and /ɛ, ae/ each map on to a single vowel in Korean [19] and were more susceptible to be confused than others for Korean speakers as shown in Figure 1 (58% for /i/, 33% for /ɪ/, 16% for /ɛ/ and 37% for /ae/). The confusion matrices for 9 vowel categories from the forward mapping model. ...
Similar publications
The current research is about the phenomenon of lexical borrowing of English vocabulary items in Punjabi language. The corpus of two million (2M) words of Punjabi language was collected and frequency list of borrowed words was developed. The frequency list of borrowed words was manually checked and the words borrowed from English were extracted. Th...
