| 研究課題/領域番号 |
20K12080
|
|---|
| 研究機関 | 会津大学 |
|---|
研究代表者 |
MARKOV K 会津大学, コンピュータ理工学部, 教授 (80394998)
|
|---|
| 研究分担者 |
松井 知子 統計数理研究所, モデリング研究系, 教授 (10370090)
齋藤 純平 福島県立医科大学, 医学部, 講師 (50332929)
|
|---|
| 研究期間 (年度) |
2020-04-01 – 2023-03-31
|
| キーワード | cough recognition / cough monitoring / DNN audio analysis |
|---|
| 研究実績の概要 |
In the fiscal 2021 year, we continued our research on cough detection using better and more sophisticated methods. Previously, we utilized several techniques for sound recognition and detection used mainly in speech recognition research, such as GMM and Viterbi decoding as well as simple Recurrent neural networks RNN where the input vectors were derived from MFCC analysis. Currently, we moved to state-of-the-art signal processing approached such as log-spectrogram or even signal level based convolutional networks for extracting audio features instead of the MFCC. This approach has been proven to achieve better performance in many audio classification and detection tasks. We build a baseline system which is capable of classifying audio segments into two classes using the COUGHVID database.
|
| 現在までの達成度 (区分) |
現在までの達成度 (区分)
3: やや遅れている
理由
Currently, we are working on implementing a detection mode of our classification system. The main difference is that in the classification model, the audio segment boundaries are known while in the detection mode, those boundaries are unknown and should be determined automatically. This is possible by using a sliding window over the (long) input signal and processing the data only within the window. Such approach introduces boundary identification errors which should be taken into account during the system evaluation. There is a slight delay in our research schedule due to the delay in collecting of real-world cough data at the Fukushima Medical University (FMU). However, those data are almost ready for research usage and we are anticipating to start working with them pretty soon.
|
| 今後の研究の推進方策 |
Our plans for the last year of this project include finishing the cough detection system using the newly collected data from FMU and improving its capabilities by building a special block which is capable of recognizing the pulmonary condition of the patients by estimating the cough type, i.e. Asthma, COPD, obstructive or non-obstructive. Depending on the amount of data available for each cough type, we plan to use some audio augmentation techniques in order to unify the number of audio samples per cough type. Finally, we plan to publish our findings in a journal paper where we compare the results achieved by our system with those from other studies concerning cough detection and classification tasks.
|
| 次年度使用額が生じた理由 |
Because of some delays in authorization of the data collection, a small amount was left and is to be used next year for the same purposes.
|
