Timeline for How to deal with overlap in STFT when considering the highest energy among multiple channels
Current License: CC BY-SA 4.0
26 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| S Apr 16, 2019 at 13:29 | history | bounty ended | havakok | ||
| S Apr 16, 2019 at 13:29 | history | notice removed | havakok | ||
| Apr 16, 2019 at 13:29 | vote | accept | havakok | ||
| Apr 16, 2019 at 12:07 | answer | added | Max | timeline score: 1 | |
| Apr 16, 2019 at 11:25 | comment | added | havakok | I want to extract the channel with the highest energy source per time frame to see how does the network in the baseline performers from this single channel alone. I guess you can say this is a better SNR. | |
| Apr 16, 2019 at 11:19 | comment | added | Max | What do you need the single file for? Is it just for classification purposes and you want to get best SNR? Then the solution to your problem might be quite simple. | |
| Apr 16, 2019 at 11:11 | comment | added | Max | There is a mistake in the description (first link you posted). It's stated that the four mike positions have $r=42\text{cm}$ where it's actually $r=42\text{mm}$, as stated further down. I was confused at first. | |
| Apr 16, 2019 at 10:30 | comment | added | havakok | They were simulated as a circular microphone array. You can take a look at the dataset site here or here | |
| Apr 16, 2019 at 8:50 | comment | added | Max | I guess the channels are correlated? How were they recorded? Are they time-aligned? If they were recorded by microphones, what where there positions, geometrically? | |
| Apr 14, 2019 at 8:31 | comment | added | havakok | Sorry, I meant to say bin#0 is the DC and bins #1-125 are the other 1024 bins. | |
| Apr 14, 2019 at 7:58 | comment | added | robert bristow-johnson | what's bin #0 ? | |
| Apr 14, 2019 at 7:48 | comment | added | havakok | No. The DC component is bin #1025 in my tensor. | |
| Apr 14, 2019 at 7:40 | comment | added | robert bristow-johnson | are you including DC (bin #0) and Nyquist (bin #1024) in your tensor? if your input is real, DC and Nyquist have no imaginary part and the only phase they can have is 0 or $\pm\pi$. | |
| Apr 14, 2019 at 7:38 | comment | added | havakok | Yep, I did. My bad.. | |
| Apr 14, 2019 at 7:37 | comment | added | robert bristow-johnson | you changed your $M$. it makes sense now. before you had 1024 positive frequency bins in the FFT result, but $M$=512. now $M$=1024 and it makes perfect sense now. | |
| Apr 14, 2019 at 7:36 | comment | added | havakok | I am not sure what you mean by adding two adjacent FFT bins. I am using a predefined function on the whole audio file. I did not manually implement the FFT and I am not sure what is going on inside the implementation. | |
| Apr 14, 2019 at 7:34 | history | edited | havakok | CC BY-SA 4.0 |
added 1 character in body
|
| Apr 14, 2019 at 7:30 | history | edited | robert bristow-johnson | CC BY-SA 4.0 |
added 29 characters in body
|
| Apr 14, 2019 at 7:30 | comment | added | havakok | @robertbristow-johnson I have edited with M=1024. As the documentation of librosa.core.stft states, I am zero padding the window length to match the n_fft length. | |
| Apr 14, 2019 at 7:27 | history | edited | havakok | CC BY-SA 4.0 |
added 7 characters in body
|
| Apr 14, 2019 at 7:24 | history | edited | robert bristow-johnson | CC BY-SA 4.0 |
i think the `stuff` notation is for code or symbols of code inside the text.
|
| Apr 14, 2019 at 7:22 | comment | added | robert bristow-johnson | your window length is 1920 samples and the FFT is 2048. does that mean that you are zero-padding with 128 zeros for each frame? and how big is $M$? is $M=1023$? or are you having a fewer number of bands? | |
| S Apr 14, 2019 at 7:06 | history | bounty started | havakok | ||
| S Apr 14, 2019 at 7:06 | history | notice added | havakok | Draw attention | |
| Apr 14, 2019 at 7:05 | history | edited | havakok | CC BY-SA 4.0 |
edited title
|
| Apr 11, 2019 at 7:34 | history | asked | havakok | CC BY-SA 4.0 |