DiscStich

Identifying, Aligning and Mixing Similar Audio With Variable Speed

Joren Six - IPEM, Ghent University - Belgium - joren.six@ugent.be
Feb 2023

Context

Broadcasting companies have digitized many laquer disc recordings.
  • 1930-1960
  • Short duration (+- 10 min)
  • Large archives
SIRDUKE - Saphir Innovatively Rescues VRT Disks Using Knowledge and Equipment - Project report

Context

Context

Context

Automatically make material originating from an album of discs fit for reuse.

Problems

  1. Identification: Find partial duplicates in large collections (via meta-data or overlap)
  2. Alignment: precisely align audio and find relative audio speed.
  3. Mixing: automatically propose a mixing point.

1. Identification


Amounts to finding partial duplicates in a large collection.

This problem is solved with modern acoustic fingerprinting techniques.
Six, J. (2022). Panako: a scalable audio search system. Journal of open source software, 7(78).

2. Alignment


Precise alignment of overlapping audio.

Dynamic Time Warping or cross correlation not efficient for audio with different speeds.
Müller, M. (2007). Dynamic time warping. Information retrieval for music and motion, 69-84.

2. Alignment

Fig: Extracting spectral peaks and using near-neighbors in 'ratio' space.
Beckmann, N., Kriegel, H. P., Schneider, R., & Seeger, B. (1990). The R*-tree: An efficient and robust access method for points and rectangles. In Proceedings of the 1990 ACM SIGMOD international conference on Management of data (pp. 322-331).

2. Alignment

Fig: A match shows up as a near diagonal.

2. Alignment

Fig: A command line implementation which can be called from Sonic Lineup Fig: A browser based implementation

2. Alignment - Evaluation

Chopping an audio recording and stitching it back together and evaluating the duration.

  • No speed change
              mean=0.1ms, stdev=1.2ms, N=80
  • Speed change
              mean=1.3ms, stdev=237.2ms, N=80

3. Mixing

Once alignment is done, mixing left to expert or archivist

Conclusion

DiscStich offers a way to speed up restoration of audio recordings by identifying, aligning and mixing digitized discs.

The main contribution is the audio-alignment algorithm.

Further reading

0110.be
  • Six, J. (2022). DiscStitch: towards audio-to-audio alignment with robustness to playback speed variabilities. In ISMIR 2022. International Society for Music Information Retrieval.
  • Six, J. (2020). OLAF: Overly lightweight acoustic fingerprinting. In 21st International Society for Music Information Retrieval Conference (ISMIR 2020).
  • Six, J. (2022). Panako: a scalable audio search system. Journal of open source software, 7(78).
  • Chenot, J. H., Laborelli, L., & Noiré, J. É. (2018). Saphir: optical playback of damaged and delaminated analogue audio disc records. Journal on Computing and Cultural Heritage, 11(3), 14-1.
  • Müller, M. (2007). Dynamic time warping. Information retrieval for music and motion, 69-84.
  • SIRDUKE - Saphir Innovatively Rescues VRT Disks Using Knowledge and Equipment - Project report