Representative verbal skill fragment extraction device and method based on seat voice segmentation

Abstract

The invention provides a representative verbal skill fragment extraction device and method based on seat voice segmentation, which are used for extracting a most representative voice fragment from seat voice as a verbal skill fragment, and are characterized by comprising a seat voice acquisition part for acquiring a plurality of seat voices to be processed; the verbal skill label prediction part is used for predicting each seat voice in sequence based on a preset verbal skill prediction model and outputting a verbal skill label and confidence of the seat voice; the voice sample classificationpart is used for classifying the seat voice into a representative sample set according to the verbal skill label and the confidence coefficient; a speech feature extraction unit that extracts MFCC features for each representative sample; a cut point acquisition unit that acquires a cut point for each representative sample; the voice segment cutting part is used for cutting each representative sample according to the cutting points to form voice segments; a segment feature extraction unit for extracting time-domain features and frequency-domain features from the speech segment; and a verbal skill segment acquisition unit for extracting verbal skill segments using a heuristic algorithm.

Description

technical field [0001] The invention belongs to the field of speech recognition, in particular to a device and method for extracting representative speech segments based on agent speech segmentation. Background technique [0002] The (intelligent) agent-based voice marketing model is becoming more and more popular. One of the advantages of this type of marketing model is that after mastering (potential) customer information, it can take the initiative to attack, thereby greatly exploring and utilizing customer resources. Therefore, in some agent positions such as call centers and telephone customer service, many agents often provide customers with call and answer services, so as to provide customers with services such as product introduction, consultation, and after-sales return visits. [0003] After a list of potential customers is given, whether the agent can successfully attract customers to place an order / in-store experience depends largely on the quality of the agent's...

AI Technical Summary

Problems solved by technology

However, the speech quality of all kinds of recordings is usually not good or bad, and the agent needs to spend a lot of time listening to and distinguishing various recordings, so the efficiency of speech improvement will be very low

[0005] Even if these agent recordings are marked in advance (i.e., quality scores are performed in advance), so that the agent can purposely filter out recordings of different qualities, but for each recording, the agent still needs to listen to the recording completely (or listened to most)

At the same time, agents also generate a large number of recordings every day, so learning these recordings is very time-consuming and labor-intensive, and it is difficult to serve as a reference for agents to improve the quality of their speaking skills

Images