A subjective evaluation on mixtures of crowdsourced audio recordings

Abstract

Abstract-Exploiting correlations in the audio, several works in the past have demonstrated the ability to automatically match and synchronize User Generated Recordings (UGRs) of the same event. Considering a small number of synchronized UGRs, we formulate in this paper simple linear audio mixing approaches to combine the available audio content. We use data from two different public events to perform a comparative listening test with the goal to assess the potential of such mixtures in improving the listening experience of the captured event, as opposed to when each UGR is consumed individually. The results of the listening tests indicate that, even with just a small number of overlapping UGRs, the outcome of the mixing process gains higher preference in comparison to original UGRs played back individually.

Normalization of Partly Overlapping Audio Recordings from the Same Event Based on Relative Signal Powers

“…The problem is that, if the two initial recordings have considerably different signal levels, then the resulting audio stream will be characterized by a sudden level transition at t = 7 s. However, jumps in audio level are well known as sources of irritation for the listeners [12]. Moreover, for some collaborative audio production approaches which have been recently proposed [8,11], it is essential that the instantaneous differences in the signal powers of the different mixture components are small. This motivates the use of some sort of scaling for minimizing the signal level differences between different recordings and below, we present two such candidate approaches.…”

“…The potential is particularly interesting with respect to the audio modality, as a multitude of synchronized UGRs may be utilized as a multichannel recording of the public event. As shown in [8], simple forms of combination of the different sources of audio content, such as signal superposition and stereo panning, may significantly improve the Fig. 1.…”

Maximum component elimination in mixing of user generated audio recordings

“…This may to some degree prevent, for example, recordings which are acquired at a small distance from the main acoustic sources to mask those which are acquired at distances further apart. Similar to the approach in [8], normalization is accomplished by obtaining an estimation of the average power of the signal, estimated across the entire duration of each UGR. In particular, if we let x m [n] symbolize the nth sample of the mth UGR and if N m is its duration in samples, a normalized version is obtained through the procesŝ…”

“…In [8] it was shown that even with a small number of overlapping UGRs, a monophonic mixture produced as the result of such a simple signal superposition may substantially improve the listening experience of the user, improving certain facts related both to the quality of the signal and to the quality of the content. While this approach can be easily extended for the case of stereophonic reproduction [8], in this paper we limit the applications to a monophonic setting.…”

“…The potential is particularly interesting with respect to the audio modality, as a multitude of synchronized UGRs may be utilized in order to provide a multichannel recording of the captured acoustic event. As shown in [8], simple forms of combination of the different sources of audio content, such as signal superposition and stereo panning, may significantly improve the user experience as opposed to when original UGRs are consumed individually. Using more advanced processing, Kim et al in [9], [11] demonstrate an approach for processing the synchronized audio streams in a collaborative fashion, so as to enhance the audio components which are common within the different recordings, and to suppress unwanted noise and interference which is unique in each audio capture.…”