{
  "id": 121455,
  "title": "Face Reconstruction from Voice using Generative Adversarial Networks",
  "url": "/competitions/deepfake-detection-challenge/discussion/121455",
  "author_name": "Bojan Tunguz",
  "post_date": "2019-12-13T12:23:52.881000",
  "votes": 23,
  "comment_count": 6,
  "views": 0,
  "content": "<p>Juts came across the following paper from NeurIPS 2019: </p>\n\n<p><a href=\"https://arxiv.org/pdf/1905.10604.pdf\">https://arxiv.org/pdf/1905.10604.pdf</a></p>\n\n<p>Here is the code: </p>\n\n<p><a href=\"https://github.com/cmu-mlsp/reconstructing_faces_from_voices\">https://github.com/cmu-mlsp/reconstructing_faces_from_voices</a></p>\n\n<p>In this competition some of the fake videos have voice swapped in, and not faces, so I am thinking that an algorithm that could reconstruct a face given the voice could potenitaly be useful. :) </p>",
  "messages": [
    {
      "id": 694312,
      "postDate": "2019-12-13T12:23:52.880Z",
      "content": "<p>Juts came across the following paper from NeurIPS 2019: </p>\n\n<p><a href=\"https://arxiv.org/pdf/1905.10604.pdf\">https://arxiv.org/pdf/1905.10604.pdf</a></p>\n\n<p>Here is the code: </p>\n\n<p><a href=\"https://github.com/cmu-mlsp/reconstructing_faces_from_voices\">https://github.com/cmu-mlsp/reconstructing_faces_from_voices</a></p>\n\n<p>In this competition some of the fake videos have voice swapped in, and not faces, so I am thinking that an algorithm that could reconstruct a face given the voice could potenitaly be useful. :) </p>",
      "rawMarkdown": "Juts came across the following paper from NeurIPS 2019: \n\nhttps://arxiv.org/pdf/1905.10604.pdf\n\nHere is the code: \n\nhttps://github.com/cmu-mlsp/reconstructing_faces_from_voices\n\nIn this competition some of the fake videos have voice swapped in, and not faces, so I am thinking that an algorithm that could reconstruct a face given the voice could potenitaly be useful. :) ",
      "votes": 23
    },
    {
      "id": 694864,
      "postDate": "2019-12-14T08:13:10.440Z",
      "content": "<p>This topic looks really interesting <a href=\"/tunguz\">@tunguz</a> 👍 . I m also adding a source that may be useful in researching this field:\n- <a href=\"https://speech2face.github.io/\">https://speech2face.github.io/</a> + <a href=\"https://speech2face.github.io/supplemental/index.html#\">https://speech2face.github.io/supplemental/index.html#</a></p>",
      "rawMarkdown": "This topic looks really interesting @tunguz 👍 . I m also adding a source that may be useful in researching this field:\n- https://speech2face.github.io/ + https://speech2face.github.io/supplemental/index.html#",
      "votes": 3
    },
    {
      "id": 695240,
      "postDate": "2019-12-14T20:32:39.033Z",
      "content": "<p>There is an ethical aspect to the subject too, some posters were kinda protested quietly due to a possibility of discriminative results. It's also very sensitive subject. </p>\n\n<p>But still people researching this area, e.g. <a href=\"https://speech2face.github.io\">https://speech2face.github.io</a>.  This one is from MIT.</p>\n\n<p>Edit: I didn't see <a href=\"/neongen\">@neongen</a> 's comment, sorry. I posted the same link after 12 hours</p>",
      "rawMarkdown": "There is an ethical aspect to the subject too, some posters were kinda protested quietly due to a possibility of discriminative results. It's also very sensitive subject. \n\nBut still people researching this area, e.g. https://speech2face.github.io.  This one is from MIT.\n\nEdit: I didn't see @neongen 's comment, sorry. I posted the same link after 12 hours\n\n"
    },
    {
      "id": 696108,
      "postDate": "2019-12-16T06:03:41.563Z",
      "content": "<p>Genius idea</p>",
      "rawMarkdown": "Genius idea",
      "votes": 1
    },
    {
      "id": 733636,
      "postDate": "2020-01-31T11:25:34.053Z",
      "content": "<p>reconstructing faces from voices, Good idea, very helpful</p>",
      "rawMarkdown": "reconstructing faces from voices, Good idea, very helpful"
    },
    {
      "id": 694902,
      "postDate": "2019-12-14T10:12:52.397Z",
      "rawMarkdown": "",
      "votes": -2,
      "isDeleted": true,
      "replies": [
        {
          "id": 3187337,
          "postDate": "2025-04-25T21:15:55.520Z",
          "content": "<p>Your brain automatically, and at great processing cost, creates a profile of each new voice you hear. Some of the factors your brain considers is gender, age, body size, race, smoking, emotional state, body position (standing, sitting), health, intelligence, etc. </p>\n<p>Creating AI faces to go with voices might be a way to challenge our innate prejudices and stereotypes. Like all tech, it is morally ambiguous, with benefits and harms. </p>\n<p>A good book to start with for this topic is \"Wired for Speech: How Voice Activates and Advances the Human-Computer Relationship\" by Clifford Nass and Scott Brave, 2007. Old, but a good entry point. </p>",
          "rawMarkdown": "Your brain automatically, and at great processing cost, creates a profile of each new voice you hear. Some of the factors your brain considers is gender, age, body size, race, smoking, emotional state, body position (standing, sitting), health, intelligence, etc. \n\nCreating AI faces to go with voices might be a way to challenge our innate prejudices and stereotypes. Like all tech, it is morally ambiguous, with benefits and harms. \n\nA good book to start with for this topic is \"Wired for Speech: How Voice Activates and Advances the Human-Computer Relationship\" by Clifford Nass and Scott Brave, 2007. Old, but a good entry point. "
        }
      ]
    }
  ],
  "comments": [
    {
      "id": 694864,
      "author_name": "neongen",
      "author_url": "",
      "post_date": "2019-12-14T08:13:10.440000",
      "content": "<p>This topic looks really interesting <a href=\"/tunguz\">@tunguz</a> 👍 . I m also adding a source that may be useful in researching this field:\n- <a href=\"https://speech2face.github.io/\">https://speech2face.github.io/</a> + <a href=\"https://speech2face.github.io/supplemental/index.html#\">https://speech2face.github.io/supplemental/index.html#</a></p>",
      "votes": 3,
      "replies": []
    },
    {
      "id": 695240,
      "author_name": "Firat Gonen",
      "author_url": "",
      "post_date": "2019-12-14T20:32:39.033000",
      "content": "<p>There is an ethical aspect to the subject too, some posters were kinda protested quietly due to a possibility of discriminative results. It's also very sensitive subject. </p>\n\n<p>But still people researching this area, e.g. <a href=\"https://speech2face.github.io\">https://speech2face.github.io</a>.  This one is from MIT.</p>\n\n<p>Edit: I didn't see <a href=\"/neongen\">@neongen</a> 's comment, sorry. I posted the same link after 12 hours</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 696108,
      "author_name": "jean9113",
      "author_url": "",
      "post_date": "2019-12-16T06:03:41.563000",
      "content": "<p>Genius idea</p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 733636,
      "author_name": "beeaware (looking for job)",
      "author_url": "",
      "post_date": "2020-01-31T11:25:34.053000",
      "content": "<p>reconstructing faces from voices, Good idea, very helpful</p>",
      "votes": 0,
      "replies": []
    },
    {
      "id": 694902,
      "author_name": "",
      "author_url": "",
      "post_date": "2019-12-14T10:12:52.397000",
      "content": "",
      "votes": -2,
      "replies": [
        {
          "id": 3187337,
          "author_name": "Fuzzball",
          "author_url": "",
          "post_date": "2025-04-25T21:15:55.520000",
          "content": "<p>Your brain automatically, and at great processing cost, creates a profile of each new voice you hear. Some of the factors your brain considers is gender, age, body size, race, smoking, emotional state, body position (standing, sitting), health, intelligence, etc. </p>\n<p>Creating AI faces to go with voices might be a way to challenge our innate prejudices and stereotypes. Like all tech, it is morally ambiguous, with benefits and harms. </p>\n<p>A good book to start with for this topic is \"Wired for Speech: How Voice Activates and Advances the Human-Computer Relationship\" by Clifford Nass and Scott Brave, 2007. Old, but a good entry point. </p>",
          "votes": 0,
          "replies": []
        }
      ]
    }
  ],
  "raw_markdown_by_id": {
    "694312": "Juts came across the following paper from NeurIPS 2019: \n\nhttps://arxiv.org/pdf/1905.10604.pdf\n\nHere is the code: \n\nhttps://github.com/cmu-mlsp/reconstructing_faces_from_voices\n\nIn this competition some of the fake videos have voice swapped in, and not faces, so I am thinking that an algorithm that could reconstruct a face given the voice could potenitaly be useful. :) ",
    "694864": "This topic looks really interesting @tunguz 👍 . I m also adding a source that may be useful in researching this field:\n- https://speech2face.github.io/ + https://speech2face.github.io/supplemental/index.html#",
    "695240": "There is an ethical aspect to the subject too, some posters were kinda protested quietly due to a possibility of discriminative results. It's also very sensitive subject. \n\nBut still people researching this area, e.g. https://speech2face.github.io.  This one is from MIT.\n\nEdit: I didn't see @neongen 's comment, sorry. I posted the same link after 12 hours\n\n",
    "696108": "Genius idea",
    "733636": "reconstructing faces from voices, Good idea, very helpful",
    "694902": ""
  }
}