Skip to main content
EURASIP Journal on Audio, Speech, and Music Processing

Table 4 Compared with other work performance on RT09. Scoring overlapped speech is accounted in the error rates

From: Latent class model with application to speaker diarization

Works

Approaches

Given speaker #

VAD[%]

Miss[%]

FA[%]

SE[%]

DER[%]

[54]

aIB

No

–

11.6

1.1

14.3

27.0

[31]

GMM+BIC

No

2.7

–

–

8.7

18.0

[32]

BottomUp

No

5.9

–

–

–

31.3

[25]

TopDown

No

–

–

–

–

21.1

[40]

BottomUp+TopDown

No

9.0

–

–

8.8

17.8

Ours

LCM

Yes

–

8.0

3.8

5.9

17.8

Back to article page