Strange results about baseline

[dai-lp]

Hi! I made only a few changes to the code to adapt it to the environment and then ran the training program to get the results shown in the figure, getting many predictions on non-existent speech events, which resulted in low accuracy and error rates, but other metrics were normal.

[dai-lp]

Class-wise metrics

Event label  | Nref    Nsys  | F        Pre      Rec    | ER       Del      Ins    | Sens     Spec     Bacc     Acc     
------------ | -----   ----- | ------   ------   ------ | ------   ------   ------ | ------   ------   ------   ------  
metro leav.. | 0       1390  | nan%     0.0%     nan%   | 6260..   0.00     6260.. | 0.0%     87.6%    43.8%    87.6%   
birds_sing.. | 0       2517  | nan%     0.0%     nan%   | 1133..   0.00     1133.. | 0.0%     77.5%    38.7%    77.5%   
coffee mac.. | 0       297   | nan%     0.0%     nan%   | 1337..   0.00     1337.. | 0.0%     97.3%    48.7%    97.3%   
furniture .. | 0       613   | nan%     0.0%     nan%   | 2760..   0.00     2760.. | 0.0%     94.5%    47.3%    94.5%   
car          | 0       2584  | nan%     0.0%     nan%   | 1163..   0.00     1163.. | 0.0%     76.9%    38.4%    76.9%   
metro appr.. | 0       1784  | nan%     0.0%     nan%   | 8034..   0.00     8034.. | 0.0%     84.1%    42.0%    84.1%   
large_vehi.. | 0       649   | nan%     0.0%     nan%   | 2922..   0.00     2922.. | 0.0%     94.2%    47.1%    94.2%   
Dishes       | 754     726   | 73.6%    75.1%    72.3%  | 0.52     0.28     0.24   | 72.3%    98.3%    85.3%    96.5%   
Cat          | 728     783   | 86.6%    83.5%    89.8%  | 0.28     0.10     0.18   | 89.8%    98.8%    94.3%    98.2%   
Vacuum_cle.. | 801     827   | 90.9%    89.5%    92.4%  | 0.18     0.08     0.11   | 92.4%    99.2%    95.8%    98.7%   
Speech       | 3745    3781  | 93.7%    93.3%    94.2%  | 0.13     0.06     0.07   | 94.2%    96.6%    95.4%    95.8%   
Electric_s.. | 522     645   | 86.4%    78.1%    96.6%  | 0.30     0.03     0.27   | 96.6%    98.7%    97.6%    98.6%   
people tal.. | 0       5667  | nan%     0.0%     nan%   | 2552..   0.00     2552.. | 0.0%     49.3%    24.7%    49.3%   
brakes_squ.. | 0       565   | nan%     0.0%     nan%   | 2544..   0.00     2544.. | 0.0%     94.9%    47.5%    94.9%   
Running_wa.. | 1385    1486  | 88.2%    85.2%    91.4%  | 0.24     0.09     0.16   | 91.4%    97.8%    94.6%    97.0%   
wind_blowing | 0       457   | nan%     0.0%     nan%   | 2058..   0.00     2058.. | 0.0%     95.9%    48.0%    95.9%   
door opens.. | 0       462   | nan%     0.0%     nan%   | 2080..   0.00     2080.. | 0.0%     95.9%    47.9%    95.9%   
shopping c.. | 0       642   | nan%     0.0%     nan%   | 2891..   0.00     2891.. | 0.0%     94.3%    47.1%    94.3%   
footsteps    | 0       1390  | nan%     0.0%     nan%   | 6260..   0.00     6260.. | 0.0%     87.6%    43.8%    87.6%   
cutlery an.. | 0       798   | nan%     0.0%     nan%   | 3593..   0.00     3593.. | 0.0%     92.9%    46.4%    92.9%   
Blender      | 538     565   | 83.4%    81.4%    85.5%  | 0.34     0.14     0.20   | 85.5%    99.0%    92.3%    98.4%   
Frying       | 794     835   | 90.0%    87.8%    92.3%  | 0.21     0.08     0.13   | 92.3%    99.0%    95.7%    98.5%   
children v.. | 0       3464  | nan%     0.0%     nan%   | 1560..   0.00     1560.. | 0.0%     69.0%    34.5%    69.0%   
Dog          | 1131    1096  | 86.8%    88.2%    85.5%  | 0.26     0.15     0.11   | 85.5%    98.7%    92.1%    97.4%   
announcement | 0       493   | nan%     0.0%     nan%   | 2220..   0.00     2220.. | 0.0%     95.6%    47.8%    95.6%   
Alarm_bell.. | 1060    1113  | 88.9%    86.8%    91.1%  | 0.23     0.09     0.14   | 91.1%    98.5%    94.8%    97.8%   
cash regis.. | 0       413   | nan%     0.0%     nan%   | 1859..   0.00     1859.. | 0.0%     96.3%    48.2%    96.3%   

like this， Could you please answer the reason? Thank you!