Non-negative Matrix Factorization and Its Variants for Audio Signal Processing

Kameoka, Hirokazu

doi:10.1007/978-4-431-55387-8_2

Cited by 16 publications

(12 citation statements)

References 32 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Therefore, there are many cases where processing magnitude spectrograms can deal with problems more easily than directly processing time-domain signals. In fact, many methods for monaural audio source separation are applied to magnitude spectrograms [1][2][3]. Furthermore, a magnitude spectrogram representation was recently found to be reasonable and effective for use with speech synthesis systems [4,5].…”

Section: Introductionmentioning

confidence: 99%

Generative adversarial network-based approach to signal reconstruction from magnitude spectrogram

Oyamada

Kameoka

Kaneko

et al. 2018

2018 26th European Signal Processing Conference (EUSIPCO)

Self Cite

View full text Add to dashboard Cite

In this paper, we address the problem of reconstructing a timedomain signal (or a phase spectrogram) solely from a magnitude spectrogram. Since magnitude spectrograms do not contain phase information, we must restore or infer phase information to reconstruct a time-domain signal. One widely used approach for dealing with the signal reconstruction problem was proposed by Griffin and Lim. This method usually requires many iterations for the signal reconstruction process and depending on the inputs, it does not always produce highquality audio signals. To overcome these shortcomings, we apply a learning-based approach to the signal reconstruction problem by modeling the signal reconstruction process using a deep neural network and training it using the idea of a generative adversarial network. Experimental evaluations revealed that our method was able to reconstruct signals faster with higher quality than the Griffin-Lim method.

show abstract

Section: Introductionmentioning

confidence: 99%

Generative adversarial network-based approach to signal reconstruction from magnitude spectrogram

Oyamada

Kameoka

Kaneko

et al. 2018

2018 26th European Signal Processing Conference (EUSIPCO)

Self Cite

View full text Add to dashboard Cite

show abstract

“…First, we evaluated the effect of the proposed algorithm in a speech enhancement task, namely l ∈ {s, n}. For comparison, we tested (i) the standard supervised NMF method [21] with Euclidean distance (SNMF_EU), KL divergence (SNMF_KL), and IS divergence (SNMF_IS); (ii) DNMF using the MU-based basis training algorithm [11] with KL divergence (DNMF_MU_KL) and Euclidean distance (DNMF_MU_EU); and (iii) DNMF using the proposed basis training algorithm with KL divergence (DNMF_MM_KL) and IS divergence (DNMF_MM_IS). Note that we have excluded DNMF_MU_IS from the baselines since it was not studied in [11].…”

Section: Experimental Evaluations a Speech Enhancement Taskmentioning

confidence: 99%

Majorization-Minimization Algorithm for Discriminative Non-Negative Matrix Factorization

Kameoka

Makino

2020

IEEE Access

Self Cite

View full text Add to dashboard Cite

“…Then, we decompose Y into the matrix H with bases for musical instrument parts, and that U with temporal activation for those bases. The framework of sound source separation by NMF is formalized as minimizing the objective function in (1) [9],…”

Section: The Model Of Separating Musical Sound Sourcesmentioning

confidence: 99%

Novel approach to separation of musical signal sources by NMF

Yazawa

Hamanaka

Utsuro

2014

2014 12th International Conference on Signal Processing (ICSP)

View full text Add to dashboard Cite

This paper proposes a method to separate polyphonic music signal into signals of each musical instrument by NMF: Non-negative Matrix Factorization based on preservation of spectrum envelope. Sound source separation is taken as a fundamental issue in music signal processing and NMF is becoming common to solve it because of its versatility and compatibility with music signal processing. Our method bases on a common feature of harmonic signal: spectrum envelopes of musical signal in close pitches played by the harmonic music instrument would be similar. We estimate power spectrums of each instrument by NMF with restriction to synchronize spectrum envelope of bases which are allocated to all possible center frequencies of each instrument. This manipulation means separation of components which refers to tones of each instrument and realizes both of separation without pre-training and separation of signal including harmonic and non-harmonic sound. We had an experiment to decompose mixture sound signal of MIDI instruments into each instrument and evaluated the result by SNR of single MIDI instrument sound signals and separated signals. As a result, SNR of lead guitar and drums approximately marked 3.6 and 6.0 dB and showed significance of our method.

show abstract

Non-negative Matrix Factorization and Its Variants for Audio Signal Processing

Cited by 16 publications

References 32 publications

Generative adversarial network-based approach to signal reconstruction from magnitude spectrogram

Generative adversarial network-based approach to signal reconstruction from magnitude spectrogram

Majorization-Minimization Algorithm for Discriminative Non-Negative Matrix Factorization

Novel approach to separation of musical signal sources by NMF

Contact Info

Product

Resources

About