Fosscomm 2022

An annotation tool helps people (without the need for specific knowledge) to mark a segment of an audio file (waveform), an image or text etc. in order to specify the segment’s properties. Annotation tools are used in machine learning applications such as Natural Language Processing (NLP) and Object Detection in order to train machines to identify objects or text. While there is a variety of annotation tools, most of them lack the multi-user feature (multiple users annotating a single project simultaneously) whose implementation is planned in this project. The audio annotation process is usually tedious and time consuming therefore, these tools (annotation tools which provide the multi-user feature) are necessary in order to reduce the effort needed as well as to enhance the quality of annotations. Since in most tasks related to audio classification, speech recognition, music detection etc., machine and deep learning models are trained and evaluated with the use audio that has previously been annotated by humans, the implementation of such a tool will lead to higher accuracy of annotated files, as they will have been annotated by more than one human, providing a more reliable dataset. In effect, multi-user annotation will reduce the possibility of human error e.g. an occasional mistaken labelling of a segment might be pointed out by another annotator. Deep learning models can be used for annotation and can kickstart your development effort by enabling faster annotation of datasets for AI algorithms. Deep learning models are sensitive to the data used to train them, this makes it hard to train the deep learning models on a specific dataset and deploy them on a different dataset. As a solution, transfer learning for sound could help adapt pretrained models into various datasets. Deep learning models used for annotation can be tuned and improved by retraining these pretrained models based on new datasets.