{ "id": 121268, "title": "Fake audio detection could also help", "url": "/competitions/deepfake-detection-challenge/discussion/121268", "author_name": "", "post_date": "2019-12-12T06:13:00.835511100Z", "votes": 10, "comment_count": 6, "views": 0, "content": "

The challenge is of finding fake videos and we should also take care of the audio and voice too. Here I am sharing a link to a potentially useful papers that could help in the detection of manipulated audio too.\n Alterations In Audio Files Using Spectrograph Analysis\nFake voice detection

\n\n

And if you are looking to understand how these fake voices are generated, this is the paper that should be refered::\nTacotron

", "messages": [ { "id": "693198", "postDate": "12/12/2019 06:13:00", "content": "

The challenge is of finding fake videos and we should also take care of the audio and voice too. Here I am sharing a link to a potentially useful papers that could help in the detection of manipulated audio too.\n Alterations In Audio Files Using Spectrograph Analysis\nFake voice detection

\n\n

And if you are looking to understand how these fake voices are generated, this is the paper that should be refered::\nTacotron

", "rawMarkdown": "The challenge is of finding fake videos and we should also take care of the audio and voice too. Here I am sharing a link to a potentially useful papers that could help in the detection of manipulated audio too.\n [Alterations In Audio Files Using Spectrograph Analysis](https://www.ijstr.org/final-print/oct2016/Detection-Of-Alterations-In-Audio-Files-Using-Spectrograph-Analysis.pdf)\n[Fake voice detection](https://github.com/dessa-public/fake-voice-detection)\n\nAnd if you are looking to understand how these fake voices are generated, this is the paper that should be refered::\n[Tacotron](https://arxiv.org/abs/1703.10135)", "votes": null }, { "id": "694260", "postDate": "12/13/2019 11:03:07", "content": "

hi @thanatoz , imho you're right, the description of the challenge

\n\n
\n

Identify videos with facial or voice manipulations

\n
\n\n

but when I checked the dataset and tried basic models I realized that voice is not that important... moreover, they don't say nothing about the voice modifications at the official paper... https://www.kaggle.com/c/deepfake-detection-challenge/discussion/121441#latest-694246

", "rawMarkdown": "hi @thanatoz , imho you're right, the description of the challenge\n> Identify videos with facial or voice manipulations\n\nbut when I checked the dataset and tried basic models I realized that voice is not that important... moreover, they don't say nothing about the voice modifications at the official paper... https://www.kaggle.com/c/deepfake-detection-challenge/discussion/121441#latest-694246", "votes": null }, { "id": "694993", "postDate": "12/14/2019 12:38:18", "content": "

What if for particular video, a video is original, voice is fake and file is classed as FAKE? Is this possible ? If yes then this will introduce noise of only detecting fakes in videos

", "rawMarkdown": "What if for particular video, a video is original, voice is fake and file is classed as FAKE? Is this possible ? If yes then this will introduce noise of only detecting fakes in videos", "votes": null }, { "id": "695204", "postDate": "12/14/2019 18:53:26", "content": "

The training set 01 . Has one woman talking .Many of the fake videos she has a relaxed tone with different accent . Whereas if you see one of her original video , she speaks quite fast with a different accent . My question is , the audio seemed to be just swapped .Not generated . How do you understand that without prior knowledge ? How do you know how someone speaks ?

", "rawMarkdown": "The training set 01 . Has one woman talking .Many of the fake videos she has a relaxed tone with different accent . Whereas if you see one of her original video , she speaks quite fast with a different accent . My question is , the audio seemed to be just swapped .Not generated . How do you understand that without prior knowledge ? How do you know how someone speaks ?", "votes": null }, { "id": "699190", "postDate": "12/20/2019 06:16:24", "content": "

Did you guys happen to go through this discussion: https://www.kaggle.com/c/deepfake-detection-challenge/discussion/121861 by @ryches ? He has done some analysis on the data for voice manipulations and found ~8% forged audio.

", "rawMarkdown": "Did you guys happen to go through this discussion: https://www.kaggle.com/c/deepfake-detection-challenge/discussion/121861 by @ryches ? He has done some analysis on the data for voice manipulations and found ~8% forged audio.", "votes": null }, { "id": "700966", "postDate": "12/22/2019 22:09:41", "content": "

I took a shot at labeling the clips with faked audio. I talk about my findings a bit here: https://www.kaggle.com/c/deepfake-detection-challenge/discussion/122785

", "rawMarkdown": "I took a shot at labeling the clips with faked audio. I talk about my findings a bit here: https://www.kaggle.com/c/deepfake-detection-challenge/discussion/122785", "votes": null }, { "id": "701151", "postDate": "12/23/2019 06:47:18", "content": "

Amazing work @ke8ctn.

", "rawMarkdown": "Amazing work @ke8ctn.", "votes": null } ], "comments": [ { "id": 694260, "author_name": "jesucristo", "author_url": "", "post_date": "12/13/2019 11:03:07", "content": "

hi @thanatoz , imho you're right, the description of the challenge

\n\n
\n

Identify videos with facial or voice manipulations

\n
\n\n

but when I checked the dataset and tried basic models I realized that voice is not that important... moreover, they don't say nothing about the voice modifications at the official paper... https://www.kaggle.com/c/deepfake-detection-challenge/discussion/121441#latest-694246

", "votes": null, "replies": [ { "id": 695204, "author_name": "phoenix9032", "author_url": "", "post_date": "12/14/2019 18:53:26", "content": "

The training set 01 . Has one woman talking .Many of the fake videos she has a relaxed tone with different accent . Whereas if you see one of her original video , she speaks quite fast with a different accent . My question is , the audio seemed to be just swapped .Not generated . How do you understand that without prior knowledge ? How do you know how someone speaks ?

", "votes": null, "replies": [] }, { "id": 699190, "author_name": "thanatoz", "author_url": "", "post_date": "12/20/2019 06:16:24", "content": "

Did you guys happen to go through this discussion: https://www.kaggle.com/c/deepfake-detection-challenge/discussion/121861 by @ryches ? He has done some analysis on the data for voice manipulations and found ~8% forged audio.

", "votes": null, "replies": [] } ] }, { "id": 694993, "author_name": "miwojc", "author_url": "", "post_date": "12/14/2019 12:38:18", "content": "

What if for particular video, a video is original, voice is fake and file is classed as FAKE? Is this possible ? If yes then this will introduce noise of only detecting fakes in videos

", "votes": null, "replies": [] }, { "id": 700966, "author_name": "ke8ctn", "author_url": "", "post_date": "12/22/2019 22:09:41", "content": "

I took a shot at labeling the clips with faked audio. I talk about my findings a bit here: https://www.kaggle.com/c/deepfake-detection-challenge/discussion/122785

", "votes": null, "replies": [ { "id": 701151, "author_name": "thanatoz", "author_url": "", "post_date": "12/23/2019 06:47:18", "content": "

Amazing work @ke8ctn.

", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "693198": "The challenge is of finding fake videos and we should also take care of the audio and voice too. Here I am sharing a link to a potentially useful papers that could help in the detection of manipulated audio too.\n [Alterations In Audio Files Using Spectrograph Analysis](https://www.ijstr.org/final-print/oct2016/Detection-Of-Alterations-In-Audio-Files-Using-Spectrograph-Analysis.pdf)\n[Fake voice detection](https://github.com/dessa-public/fake-voice-detection)\n\nAnd if you are looking to understand how these fake voices are generated, this is the paper that should be refered::\n[Tacotron](https://arxiv.org/abs/1703.10135)", "694260": "hi @thanatoz , imho you're right, the description of the challenge\n> Identify videos with facial or voice manipulations\n\nbut when I checked the dataset and tried basic models I realized that voice is not that important... moreover, they don't say nothing about the voice modifications at the official paper... https://www.kaggle.com/c/deepfake-detection-challenge/discussion/121441#latest-694246", "694993": "What if for particular video, a video is original, voice is fake and file is classed as FAKE? Is this possible ? If yes then this will introduce noise of only detecting fakes in videos", "695204": "The training set 01 . Has one woman talking .Many of the fake videos she has a relaxed tone with different accent . Whereas if you see one of her original video , she speaks quite fast with a different accent . My question is , the audio seemed to be just swapped .Not generated . How do you understand that without prior knowledge ? How do you know how someone speaks ?", "699190": "Did you guys happen to go through this discussion: https://www.kaggle.com/c/deepfake-detection-challenge/discussion/121861 by @ryches ? He has done some analysis on the data for voice manipulations and found ~8% forged audio.", "700966": "I took a shot at labeling the clips with faked audio. I talk about my findings a bit here: https://www.kaggle.com/c/deepfake-detection-challenge/discussion/122785", "701151": "Amazing work @ke8ctn." }, "source": "meta" }