Nonnegative Matrix Factorization
read more
Citations
Nonnegative Matrix Factorization: A Comprehensive Review
Nonnegative Matrix Factorization for Interactive Topic Modeling and Document Clustering
Past review, current progress, and challenges ahead on the cocktail party problem
Discriminative Nonnegative Matrix Factorization for dimensionality reduction
Relevance of polynomial matrix decompositions to broadband blind signal separation
References
Indexing by Latent Semantic Analysis
Learning the parts of objects by non-negative matrix factorization
Learning parts of objects by non-negative matrix factorization
Top 10 algorithms in data mining
Positive matrix factorization: A non-negative factor model with optimal utilization of error estimates of data values†
Related Papers (5)
Incremental nonnegative matrix factorization based on correlation and graph regularization for matrix completion
Convex nonnegative matrix factorization with manifold regularization
Frequently Asked Questions (11)
Q2. What is the main purpose of the system?
the system could be used as annotation tool: it could assist semi-automatic transcription music signals into musical scores, where an automatic system would infer note boundaries, rhythms, key and time signature from the user inputs.
Q3. What is the purpose of the method?
Assuming the desired source generates smooth melody lines, the melody path is then tracked in H̃F0 with a Viterbi algorithm [4]: the user-defined regions are therefore used to restrict the melody tracking.
Q4. What is the underlying mechanism of the audio mixture?
The audio mixture is modelled through its F × N short-term power spectrum (STPS) matrix S, defined as the power of its STFT X, with F the number of Fourier frequencies and N the number of frames.
Q5. What is the purpose of the study?
In order to evaluate the usage and the performance of the proposed user-guided source separation system, the development set (5 excerpts) for the SiSEC 2011 “Professionally Produced Music Recordings” task [10] is used.
Q6. How can a user choose the source?
Expert users can be asked to choose the desired source through its position [14] or selecting components that are played by the desired instrument, thanks to intermediate separation results [15].
Q7. What can be done to improve the separation of a particular source?
With multi-channel signals, such as stereo signals, one can infer spatial information [2], or train models to extract specific sources, even with single-channel signals [1].
Q8. What is the purpose of the proposed system?
The proposed system delegates the source identification to the user, such that there is less ambiguity with the definition of the target source, for the system.
Q9. What is the default for the HF0 algorithm?
All the systems and users discussed in this section used the same default following parameters: K = 4, U = 577 (for 16 F0s per semitone, from 100 to 800Hz), R = 40, F = 1025 (for Fourier tranforms of size 2048, i.e. 46.44ms@44.1kHz) and with 25 iterations of the NMF algorithm.
Q10. Why is the rap song more difficult to identify?
Some songs might be more challenging, such as the rap song (dev2 fort), probably because the desired vocal signal is closer to speech than to singing voice.
Q11. What is the use of time-frequency representations?
Time-frequency representations (TFR), such as the short-term Fourier transform (STFT), are therefore required to visually identify such sources.