Background sound evaluation for information classification and ranking.
Publication Date: 2016-Apr-13
The IP.com Prior Art Database
Disclosed is a method for enhanced ranking of the speech content based on the evaluation and sentiment of the background sounds.
Page 01 of 3
Background sound evaluation for information classification and ranking .
During speech presentations, the audience's attitude toward presented content can be reflected in various background sounds and utterances. In general, average audience sentiment is expressed by the prevailing majority. One can evaluate the background sounds during the content presentation and provide additional input to the information retrieval's confidence scoring and to the sentiment content classifiers.
1. The presenter starts speech with calm voice and is at a normal level. The audience's noise level keeps increasing. Recognized tone of the background noise indicates the majority does "not agree" or "dislikes" the speech content. For example, there are many phrases and sounds such as "bow", 'oooo', and even whistles, etc.. The presenter tries to increase the level of the voice and tone of pitches to over-sound the audience. The presenter is confident and tries to deliver his points of view no matter how they are received. The system can evaluate the speech content with a higher confidence ranking in this case.
2. The presenter follows the audience's tone and does not try to emphasize his points. The presenter voice level gets lower and lower. The presenter might even skip topics just to complete the presentation faster to avoid additional embarrassment. The system can evaluate the speech content with lower confidence ranking in this case.
3. The presenter speech is accompanied by the high level of the audience noise that indicates the majority does "not agree" with the presenter. Then the presenter might change his opinion on the presented topic to make sure his speech is approved by this audience. The audience's noise changes from "not agree" to "ready to listen and accept", so the audience becomes much quieter. The audience expresses the "agree" sounds such as applauses. The system can evaluate the speech content with lower level of confidence ranking if the shift in the audience reception impacts the opinion of the presenter and forces to "flip-flop".
The speaker confidence in these use cases is used as a measure of the content validity and an input to the content scoring in the information retrieval.
In case (2) the confidence score of the speech content delivered by the search engine is downgraded.
If the speaker voice is "dying", the speaker speech content becomes filling up with "filler" and other signs of the mumbled speech words, then one can state with some probability that the delivered content can be scored with lower confidence when retrieved by the search system.
In case (1) the confidence score of the speech content delivered by the search engine is upgraded.
If the speaker continues to deliver the speech with higher voice and pitches trying "over-shout the crowd", then the system can use this data as a feedback that the presenter is confident in his content.
Please note all those thresholds and adjustment values are establishe...