As the FT embarks on a growing production of audio articles, voiced by both computers and humans, it would be beneficial to track, or even just identify an audio file that has been shared outside its original context. Citing sources at the beginning or the end of a file is insufficient; the file can easily be cropped and amended to exclude them. Audible marks every X seconds are also not an option, as they disturb the users' experience.
A preliminary research into audio watermarking has shown that it would be possible to embed images in the spread spectrum of an audio file. Editing the spectrum manually is possible with software such as Adobe Audition. It used to allow for bitmap embedding (as demonstrated in this video), but the feature has been dropped in newer versions, making it hard to automate the watermarking process.
Useful resources
- Audio Watermarking, Wikipedia
- Visual Domain Audio Watermarking
- Watermarking vs. fingerprinting, explained by Chirp