Sound event labeling and identification method adopting double Token tags

Abstract

The invention discloses a sound event labeling and recognition method adopting double Token tags, which is characterized by comprising a sound event labeling process and a sound event recognition process, and the sound event labeling process is as follows: 1-1) realizing an audio tag form; 1-2) completing all audio annotations in the data set; the identification process comprises the following steps: 2-1) constructing an audio data set; 2-2) extracting audio data preprocessing and feature; 2-3) amplifying the audio data; 2-4) building a convolutional recurrent neural network; 2-5) training the convolutional recurrent neural network learning detection model, and 2-6) using the trained detection model to identify the audio to be detected. According to the method, the accuracy rate can be guaranteed, meanwhile, the sound event recognition range is widened at a low cost, accurate sound event detection and monitoring in the living environment of people can be achieved, and therefore the method better serves for smart city construction.

Description

Technical field [0001] The present invention relates to the field of acoustic event detection, and is specifically a sound event label and identification method using a dual Token tag. Background technique [0002] In people's living environment, all kinds of sounds carry a large number of physical event information about the daily environment and among them. Sound Events Detection (SED) studies can help people know their sound scenes, identify various sound source categories, and have important practical significance. Can be applied to urban environment noise monitoring, public space safety monitoring, indoor environment, children's behavioral monitoring, such as smart cities and wisdom home scenes, such as acoustic monitoring application scenarios, can automatically detect gunshots, scream, object burning In human-machine interaction, hearing sensation, meets the various types of inspection needs in society. [0003] Acoustic event detection tasks depend on the signal processin...

AI Technical Summary

Problems solved by technology

However, strong labeling of audio data is often done through human listening and manual labeling, which requires people to maintain a high degree of attention during the listening process and use professional software to record, which is a very time-consuming and labor-intensive task. When a piece of audio is mixed with multiple types of acoustic events that overlap in time, the time and labor costs of the strong labeling task will increase exponentially

For weak labeling, this method only labels whether there is an event of interest in a piece of audio, at the cost of discarding part of the time information, and reduces the labor cost of audio dataset labeling. Correspondingly, the model obtained by training with a weakly labeled dataset cannot Predict the time information of sound events, and the recognition rate is not high

Commonly used weakly labeled datasets are: 1. Detection and Classification of Acoustic Scenes and Events (DCASE2017) Acoustic Event Detection Dataset - The advantage lies in accurate labeling, but the data set has a small number of samples and the model trained using this data set The recognition range is narrow, and the model universality is poor; 2. Google Audio-set weakly labeled data set - the advantage lies in the large number of sample types and quantities, but due to cost constraints, its labeling accuracy is low. Therefore, the training based on this data set model, although the recognition range is wider, but the recognition accuracy is not as good as the former

Images