MULAN: A Blind and Off-Grid Method for Multichannel Echo Retrieval
Authors: Helena Peic Tukuljac, Antoine Deleforge, Remi Gribonval
NeurIPS 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In section 4 the method is compared and evaluated on both synthetic and room acoustic data. We used two distinct metrics to evaluate Dirac location estimation and Dirac weight estimation. These metrics for 100 onand off-grid tests and all three methods are showed in Table 1. |
| Researcher Affiliation | Academia | Helena Pei c Tukuljac Department of Computer and Communication Sciences École polytechnique fédérale de Lausanne helena.peictukuljac@epfl.ch Antoine Deleforge Université de Lorraine, CNRS, Inria, LORIA F-54000 Nancy, France antoine.deleforge@inria.fr Rémi Gribonval Univ Rennes, Inria, CNRS, IRISA 35000 Rennes, France remi.gribonval@inria.fr |
| Pseudocode | Yes | Algorithm 1 MULAN (MULtichannel ANnihilation) |
| Open Source Code | Yes | The code for this submission can be found at: https://github.com/epfl-lts2/mulan. |
| Open Datasets | Yes | a point source emitting speech from the TIMIT dataset [31] |
| Dataset Splits | No | The paper describes experiments and evaluations but does not provide specific training, validation, or test dataset splits (percentages or sample counts). |
| Hardware Specification | No | The paper does not provide specific hardware details (like GPU/CPU models, processor types, or memory amounts) used for running the experiments. |
| Software Dependencies | No | Simulations were performed using the pyroomacoustics library [32]. (No version specified for pyroomacoustics or any other software.) |
| Experiment Setup | Yes | Simulations were performed using the pyroomacoustics library [32]. ... The absorption coefficient of each surface of the room is set to 0.2. ... The filters are simulated in the continuous-time domain using the image-source method [33]. They are then smoothed, sampled and convolved with the source signal at Fs = 16k Hz ... The M-channel input signals used are 0.25s long, i.e., N = 0.25Fs = 4000 samples. ... For MULAN, the DFT (eq. 10) is applied to each input signal using a grid F of F = 401 regularly spaced frequencies between 200 Hz and 2000 Hz. ... An odd number of frequencies was chosen... We use 20 random initializations as a good compromise between global convergence and computational complexity, max_iter= 1000 and conv_thresh= 0.1%. ... The filters lengths L were always set to the true lengths (which never exceed 0.05Fs) and the sparsity parameter λ for LASSO was manually set to λ = 10 3... RMSE thresholds were defined for success of recovery: 1 sample for locations as before and 10 2 for weights. |