List of Tables

1. DER improvement by using a speech/non-speech detector
2. Comparison of PLP12 and MFCC19 parameterizations on RT04f
3. Comparison between BIC and Viterbi stopping criterions for RT04f data
4. Estimated SNR for RT04f Broadcast News shows
5. Estimated SNR for RT06s Conference Meetings, MDM channels
6. Estimated SNR for RT06s Conference Meetings, IHM channels
7. Estimated SNR for RT06s Lecture Room Meetings, MDM channels
8. Average total speaker duration in RT06s conference room data
9. Average total speaker duration in RT06s lecture room data
10. Average total speaker duration in RT04f broadcast news data
11. Average number of speaker for Rt04f and RT06s
12. Speaker turn durations for RT04f and RT06s
13. Overlap analysis between hand and forced alignments in RT06s conference room meetings
14. Main differences between Meetings and Broadcast News recordings
15. Summary of datasets used in the experiments
16. Results for the CV-EM training algorithm in the agglomerate system
17. Speech/non-speech errors on development and test data
18. DER using different speech/non-speech systems
19. RT06s system filter-and-sum performance
20. Reference channel estimation results
21. Post-processing algorithms results
22. Results for various percentages of noise thresholding
23. Results for alternative Viterbi weights
24. Results for alternative Viterbi weights
25. Results for relative channel weights and elimination of bad channels
26. WER using RT06s ASR system including the presented beamformer
27. DER for the development and evaluation sets for models complexity selection and initial number of clusters algorithms
28. Average DER for the development set using various possible distance metrics and initial segment selection criterions
29. DER results for the segment and frame purification algorithms
30. DER results for different weight selection algorithms
31. Summary of DER improvements for the individual algorithms in the development set
32. Results for the model complexity and number of initial clusters in the agglomerate system
33. Results for the friends-and-enemies algorithm
34. Results for the CV-EM training algorithm in the agglomerate system
35. Results for the frame purification algorithm in the agglomerate system
36. Results for the segment purification algorithm in the agglomerate system
37. Summary of average DER for the agglomerate system on development and evaluation data
38. Results for the TDOA-MDM task using different algorithm settings
39. Overall thesis scores comparison
40. Overall thesis DER scores comparison
41. Systems summary description and DER on the evaluation set for RT05s
42. Results for RT06s Speaker Diarization, conference room environment
43. Results for RT06s Speaker Diarization, lecture room environment
44. Results for RT06s Speech Activity Detection (SAD). Results with * are only for a subset of segments
45. Summary of datasets used in the experiments