AI

Bitrate Classification of Twice-Encoded Audio using Objective Quality Features

Abstract

Streaming services such as Google Play Music and Sound-Cloud handle terabytes of audio data every week. These services aim to encode audio with a balance between quality of experience (QoE) [1] for the end user, the size of the encoded audio files, and the processing cost of the encoding. Users may upload files to a streaming service that have already been encoded because the user wants to reduce file size to decrease upload time. The same audio encoded as a 3 MB uncompressed WAV, a 510 KB 256kb/s AAC-LC, or a 250 KB 128 kb/s Opus all seem similar in quality to expert listeners [2]. Streaming services encode audio to a number of bitrates and formats to provide the best experience for users of different devices. For example, mobile users may prefer to compromise quality to limit bandwidth consumption. Services do not encode to bitrates higher than that of the uploaded files as there will be no increase in quality. Determining the lowest bitrate of the files allows the streaming service to forgo encoding the files to bitrates higher than that of the uploaded files, saving on processing and storage space.