1 of 11

A Dataset of Larynx Microphone Recordings�for Singing Voice Reconstruction

Simon Schwär, Michael Krause, Michael Fast, Sebastian Rosenzweig,

Frank Scherbaum, and Meinard Müller

2 of 11

© AudioLabs, 2024

Schwär et al.

A Dataset of Larynx Microphone Recordings for Singing Voice Reconstruction

2

3 of 11

Room Microphone

© AudioLabs, 2024

Schwär et al.

A Dataset of Larynx Microphone Recordings for Singing Voice Reconstruction

3

4 of 11

Close-up Microphone (CM)

© AudioLabs, 2024

Schwär et al.

A Dataset of Larynx Microphone Recordings for Singing Voice Reconstruction

4

5 of 11

© AudioLabs, 2024

Schwär et al.

A Dataset of Larynx Microphone Recordings for Singing Voice Reconstruction

5

6 of 11

Larynx Microphone (LM)

© AudioLabs, 2024

Schwär et al.

A Dataset of Larynx Microphone Recordings for Singing Voice Reconstruction

6

7 of 11

We want

cross-talk free, high-quality, in-situ recordings�of the singing voice for listening and computational analysis

We have

either high-quality CM recordings with cross-talk

or low-quality LM recordings without cross-talk

© AudioLabs, 2024

Schwär et al.

A Dataset of Larynx Microphone Recordings for Singing Voice Reconstruction

7

8 of 11

MIR Dataset

Larynx Microphone Singer-Songwriter Dataset (LM-SSD)

MIR Task

Singing Voice Reconstruction (SVR)

from Larynx Microphone Recordings

© AudioLabs, 2024

Schwär et al.

A Dataset of Larynx Microphone Recordings for Singing Voice Reconstruction

8

9 of 11

MIR Dataset

Larynx Microphone Singer-Songwriter Dataset (LM-SSD)

​

  • 12 songs, over 4 hours of recordings, 4 different singers
  • 2 different LMs, CM, separate guitar mics, guitar pickup
  • 2 mixes (with and w/o effects)
  • Systematic split of CM signals with or w/o crosstalk

© AudioLabs, 2024

Schwär et al.

A Dataset of Larynx Microphone Recordings for Singing Voice Reconstruction

9

10 of 11

LM-SVR Baseline System

Using Differentiable DSP

© AudioLabs, 2024

Schwär et al.

A Dataset of Larynx Microphone Recordings for Singing Voice Reconstruction

10

11 of 11

Simon Schwär, Michael Krause, Michael Fast, Sebastian Rosenzweig, Frank Scherbaum, and Meinard Müller.

A Dataset of Larynx Microphone Recordings for Singing Voice Reconstruction.

Transactions of the International Society for Music Information Retrieval, 7(1), 30–43, 2024.

transactions.ismir.net/articles/10.5334/tismir.166

© AudioLabs, 2024

Schwär et al.

A Dataset of Larynx Microphone Recordings for Singing Voice Reconstruction

11