![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F3004733%2F4df0b043457402ad7318d109cc5a3e80%2Fa4.png?generation=1596031716930344&alt=media\")
Data augmentation is the process by which we create new synthetic training samples by adding small perturbations on our initial training set.\nThe objective is to make our model invariant to those perturbations and enhace its ability to generalize.\nIn images data augmention can be performed by shifting the image, zooming, rotating ...\nIn our case we will add noise, stretch and roll, pitch shift ...
\n\nIt simply add some random value into data by using numpy.\n
It just shift audio to left/right with a random second.\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F3004733%2F81a5bb5f944d4484074e37df47c1ef99%2Fa1.png?generation=1596031680234729&alt=media\")
It Change pitch randomly\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F3004733%2Fe37d3982575221941d893cbbf655b32c%2Fa2.png?generation=1596031641223863&alt=media\")
It stretches times series by a fixed rate.\n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F3004733%2Fc0b6dd5bef633c0bd747eda7ed57e228%2Fa3.png?generation=1596031609456536&alt=media\")
Using these techniques we can create more data and randomness so it can generalise the results
\n\n", "messages": [ { "id": "950630", "postDate": "07/29/2020 14:03:01", "content": "\n\nData augmentation is the process by which we create new synthetic training samples by adding small perturbations on our initial training set.\nThe objective is to make our model invariant to those perturbations and enhace its ability to generalize.\nIn images data augmention can be performed by shifting the image, zooming, rotating ...\nIn our case we will add noise, stretch and roll, pitch shift ...
\n\nIt simply add some random value into data by using numpy.\n
It just shift audio to left/right with a random second.\n
It Change pitch randomly\n
It stretches times series by a fixed rate.\n
Using these techniques we can create more data and randomness so it can generalise the results
\n\n", "rawMarkdown": "[code](https://www.kaggle.com/jainarindam/audio-augmentation-methods)\n\n## Data augmentation definition :\n\nData augmentation is the process by which we create new synthetic training samples by adding small perturbations on our initial training set.\nThe objective is to make our model invariant to those perturbations and enhace its ability to generalize.\nIn images data augmention can be performed by shifting the image, zooming, rotating ...\nIn our case we will add noise, stretch and roll, pitch shift ...\n\n## 1. Adding Noise\nIt simply add some random value into data by using numpy.\n\n\n\n## 2. Shifting Time \nIt just shift audio to left/right with a random second.\n\n\n## 3. Change Pitch\nIt Change pitch randomly\n\n\n## 4. Change Speed\n It stretches times series by a fixed rate.\n\n\n### If you have more ideas please let me know in discussion I will add them\nUsing these techniques we can create more data and randomness so it can generalise the results \n\n [Check the code here](https://www.kaggle.com/jainarindam/audio-augmentation-methods)", "votes": null }, { "id": "950644", "postDate": "07/29/2020 14:14:05", "content": "I was thinking of using mirror and up-side down in Audio as we used to do in image augmentation. I am not sure if that will work. What are your thought on that?
", "rawMarkdown": "I was thinking of using mirror and up-side down in Audio as we used to do in image augmentation. I am not sure if that will work. What are your thought on that?", "votes": null }, { "id": "965075", "postDate": "08/10/2020 11:19:44", "content": "I am not sure but if I had to add another one to this it could be Volume Tuning.\nLet me know if I am right or not.
", "rawMarkdown": "I am not sure but if I had to add another one to this it could be Volume Tuning.\nLet me know if I am right or not.", "votes": null }, { "id": "965501", "postDate": "08/10/2020 17:18:52", "content": "Just like in Car :p \n![\"\"](\"https://www.googleapis.com/download/storage/v1/b/kaggle-user-content/o/inbox%2F3004733%2F5b50bc633e703c447afbe19e27571910%2Fcarrrr.jpg?generation=1597079779424528&alt=media\")
normalizing bird sound ( to make predictions for faint sounds). This is discussed here: https://www.kaggle.com/c/birdsong-recognition/discussion/171133
\nlooking forward to your implementation
Hey @jainarindam I have created a notebook with all my ideas and implemented transforms on top of albumentations library making it easy to use
\njust do albumenatations.compose
\nhttps://www.kaggle.com/tanulsingh077/audio-albumentations-transform-your-audio
\nI hope this helps
@tanulsingh077 awesome
", "rawMarkdown": "tanulsingh077 awesome", "votes": null }, { "id": "1001356", "postDate": "09/07/2020 09:19:08", "content": "Thanks for sharing Arindam. In your experience what has been the appropriate noise level to add for training ? 0.01 or something higher
", "rawMarkdown": "Thanks for sharing Arindam. In your experience what has been the appropriate noise level to add for training ? 0.01 or something higher", "votes": null }, { "id": "1001735", "postDate": "09/07/2020 14:23:26", "content": "Well I did it with higher that >0.1 got no luck. probably less should work
", "rawMarkdown": "Well I did it with higher that >0.1 got no luck. probably less should work", "votes": null }, { "id": "1001802", "postDate": "09/07/2020 15:27:09", "content": "I tried 0.03 and it didn't work for me.
", "rawMarkdown": "I tried 0.03 and it didn't work for me.", "votes": null } ], "comments": [ { "id": 972420, "author_name": "ddbishnoi", "author_url": "", "post_date": "08/16/2020 14:41:35", "content": "normalizing bird sound ( to make predictions for faint sounds). This is discussed here: https://www.kaggle.com/c/birdsong-recognition/discussion/171133
\nlooking forward to your implementation
Hey @jainarindam I have created a notebook with all my ideas and implemented transforms on top of albumentations library making it easy to use
\njust do albumenatations.compose
\nhttps://www.kaggle.com/tanulsingh077/audio-albumentations-transform-your-audio
\nI hope this helps
@tanulsingh077 awesome
", "votes": null, "replies": [] } ] }, { "id": 1001356, "author_name": "watzisname", "author_url": "", "post_date": "09/07/2020 09:19:08", "content": "Thanks for sharing Arindam. In your experience what has been the appropriate noise level to add for training ? 0.01 or something higher
", "votes": null, "replies": [ { "id": 1001735, "author_name": "jainarindam", "author_url": "", "post_date": "09/07/2020 14:23:26", "content": "Well I did it with higher that >0.1 got no luck. probably less should work
", "votes": null, "replies": [] }, { "id": 1001802, "author_name": "watzisname", "author_url": "", "post_date": "09/07/2020 15:27:09", "content": "I tried 0.03 and it didn't work for me.
", "votes": null, "replies": [] } ] }, { "id": 950644, "author_name": "jainarindam", "author_url": "", "post_date": "07/29/2020 14:14:05", "content": "I was thinking of using mirror and up-side down in Audio as we used to do in image augmentation. I am not sure if that will work. What are your thought on that?
", "votes": null, "replies": [ { "id": 965075, "author_name": "palaksood97", "author_url": "", "post_date": "08/10/2020 11:19:44", "content": "I am not sure but if I had to add another one to this it could be Volume Tuning.\nLet me know if I am right or not.
", "votes": null, "replies": [] }, { "id": 965501, "author_name": "jainarindam", "author_url": "", "post_date": "08/10/2020 17:18:52", "content": "Just like in Car :p \n