Removing overlaps on Audio recordings
- Context
- SB has a large collection of digitized radio broadcast data, approx. 350.000 hours
- These recordings are split on chunks of 2 hours (the length of the original tapes) with small overlaps between files.
- Preservation Issues
- The digitization process creates files that overlap slightly at the ends.
- To enhance user experience, SB wanted to eliminate the overlap so that the broadcast is a continuous show.
- SCAPE solution (unique selling point)
- The xCorrSound toolset allows us to detect overlaps and easily cut the recording to make it a single file/show.
- The tool can be applied in large scale (e.g. on hadoop).
- Success Story Champion
- Bolette Jurik (SB)
- Positive words:
Used in production workflows
Precision in the overlap analysis
Automated process
Leads to Improved and optimized end user experience
Solves an otherwise unsolvable problem
Innovative thinking based on well-known cross correlation
(Independent of operating system)
Resource efficient
Open source - freely available
Command line tool - easy to install and integrate into a workflow
(see description http://wiki.opf-labs.org/display/TR/xcorrSound)
Changed names for tools in xcorrSound (check xcorrSound):
waveform-compare - the former 'migrationQA', is used to compare 2 files to see if they are 'enough alike'
overlap-analysis - Is used to find the overlap between 2 files - an overlap means that the exact same section is found at the end of the first file and the start of the next.
sound-match - is used to find a short sound section every time it appears in a longer sound section. For instance, a special news jingle before and after the newscast. Sound-match can find all the newscast in a radio recording by marking the jingles.