This plot shows a clear unbalance between misses and false alarms for the classes "Speech" and "Speech over music".
The proper selection of a set of acoustic features may help to describe the behavior of the acoustic classes (speech, music, environmental sounds, etc).
Regarding the overlapped classes, speech can be found along with noise 40%% of the time and along with music 15%% of the time.
A comparison of the change scores for different classes within the same condition showed that there were no significant differences between classes, (speech alone: F 2,42) = 0.29, p = 0.746, speech plus gesture: F 2,48) = 1.05, p = 0.357).
The proposed evaluation task consisted of segmenting a broadcast news audio document into five acoustic classes: speech (SP), speech with noise (SN), speech with music (SM), music (MU), and others (OT).
The proposed method is applied to segment and classify audios coming from TV shows into five different acoustic classes: speech, music, speech with music, speech with noise, and others.
Group 3 presented a single primary system based on three independent GMM-UBM detectors of broad acoustic classes (speech, music, and noise in every possible context) with a segmentation-by-classification strategy.
This way we stimulate the participants to detect well not only the best-represented classes ("Speech" and "Speech over noise," 77% of total duration), but also the minor classes (like "Music," 5%).
In summary, the goal was to segment the incoming audio into three (possibly overlapped) acoustic classes: speech, music, and noise, where the audio was drawn from different databases that have been merged or even overlapped, thus dramatically increasing the difficulty of the task with regard to previous evaluations.
