{
 "page": "/remove-background-noise/",
 "preparedOn": "2026-10-19",
 "samples": [
  {
   "slug": "beach-wind",
   "label": "Wind and surf on a beach",
   "source": {
    "excerptStartSeconds": 57.9,
    "excerptEndSeconds": 73.0,
    "sampleRate": 48000,
    "channels": 2,
    "excerptSha256": "90da81bbf3a3b57ea645197e17595b856ec55780b90d9af85d1932a7ccf45cdb"
   },
   "restoration": {
    "app": "Diffio web app (local stack on the dev Mac mini fleet)",
    "model": "Diffio 4.5 Pro",
    "createdAtUtc": "2026-10-19T17:31:22.954Z",
    "secondsFromCreateToResult": 22.9,
    "exports": {
     "speech": {
      "name": "Restored speech",
      "sha256": "06491ad21c42596e3bb0ca4d146ca233f57000b4a9cc0345ace46311e43d1c6a"
     },
     "background": {
      "name": "Background noise",
      "sha256": "83950cd9135f57c292e6a80887e796be316677ede9864b7538bc9b9ed385afb0"
     }
    }
   },
   "playback": {
    "format": "MP3 mono 44.1 kHz 96 kb/s",
    "durationSeconds": 15.1,
    "measuredLufsBeforeGain": {
     "original": -15.2,
     "speech": -21.1
    },
    "fixedGainDb": {
     "original": -4.8,
     "speechAndBackground": 1.1
    },
    "quietestFifteenthPercentileDb": {
     "original": -24.4,
     "restoredSpeech": -40.8
    },
    "processing": "Fixed gain only. No compression, limiting or EQ."
   },
   "qualityCheck": {
    "method": "ScoreQ no-reference MOS (adapt_nr_telephone, diffio-mac bench scoreq.py and windowing.py: 10 s windows, last window end-anchored, mean) on the uploaded excerpt and the exported restored speech; higher is better. 'Leveled' is the bench's v2 loudness_normalize copy.",
    "original": 2.42,
    "restored": 4.53,
    "restoredMinWindow": 4.485,
    "improvement": 2.11,
    "originalLeveled": 2.478,
    "restoredLeveled": 4.562
   },
   "transcriptCheck": {
    "method": "faster-whisper large-v3-turbo on the page's original and restored MP3s",
    "wordMatchRatio": 0.952,
    "differences": [
     "punch -> ponds (the group's name, 'Plastic Punch')",
     "as -> us"
    ]
   }
  },
  {
   "slug": "storm-diner",
   "label": "Diner chatter during a storm",
   "source": {
    "sourcePageUrl": "https://commons.wikimedia.org/wiki/File:When_a_Hurricane_Strikes,_Weather_the_Storm_With_Waffles.webm",
    "license": "Public domain (work of Voice of America, a U.S. government agency)",
    "credit": "VOA News, September 14, 2018. A hospital dialysis worker on a break at a diner in Goldsboro, North Carolina, during Hurricane Florence",
    "excerptStartSeconds": 61.95,
    "excerptEndSeconds": 79.3,
    "sampleRate": 48000,
    "channels": 2,
    "excerptSha256": "11a45af46decfaac84423e378683de68d4c74a5249a7f1b0ab95666b7c7c2d83"
   },
   "restoration": {
    "app": "Diffio web app (local stack on the dev Mac mini fleet)",
    "model": "Diffio 4.5 Pro",
    "projectName": "Diner break interview",
    "resultSeenAtUtc": "2026-10-19T22:45:45.855Z",
    "secondsFromCreateToResult": 22.3,
    "exports": {
     "speech": {
      "name": "Restored speech",
      "sha256": "570bd491445c1f3e209428fdb0b01e5a3114412754dfeebb5b44523aa411cbdc"
     },
     "background": {
      "name": "Background noise",
      "sha256": "2915ebfd540bc6c1939d217fe68c9ab4c236fbde6a03c5e58ba8236b3a8a0619"
     }
    }
   },
   "playback": {
    "format": "MP3 mono 44.1 kHz 96 kb/s",
    "durationSeconds": 17.4,
    "measuredLufsBeforeGain": {
     "original": -19.5,
     "speech": -22.5
    },
    "fixedGainDb": {
     "original": -1.0,
     "speechAndBackground": 2.0
    },
    "quietestFifteenthPercentileDb": {
     "original": -29.9,
     "restoredSpeech": -53.4
    },
    "processing": "Fixed gain only. No compression, limiting or EQ."
   },
   "noise": "Dishes, frying and other diners talking (AudioSet classifier on the exported background: speech, dishes/pots/pans, frying, sizzle)",
   "transcriptCheck": {
    "method": "faster-whisper large-v3 (and large-v3-turbo) on the uploaded excerpt and on the restored speech, fillers (uh, um) ignored",
    "wordMatchRatio": 0.964,
    "differences": [
     "but we've -> that have"
    ],
    "note": "The page quotes the restored words. One phrase is ambiguous in the noisy original: large-v3 hears 'but we've evacuated', large-v3-turbo 'who have evacuated'; both models hear 'that have evacuated' in the restored speech."
   },
   "qualityCheck": {
    "method": "ScoreQ no-reference MOS (adapt_nr_telephone, diffio-mac bench scoreq.py and windowing.py: 10 s windows, last window end-anchored, mean) on the uploaded excerpt and the exported restored speech; higher is better. 'Leveled' is the bench's v2 loudness_normalize copy.",
    "original": 2.044,
    "restored": 4.682,
    "restoredMinWindow": 4.658,
    "improvement": 2.639,
    "originalLeveled": 2.085,
    "restoredLeveled": 4.691
   }
  },
  {
   "slug": "night-street",
   "label": "Street noise on a city street at night",
   "source": {
    "sourcePageUrl": "https://freesound.org/people/kerouacsamerica/sounds/31523/",
    "license": "CC BY 3.0, https://creativecommons.org/licenses/by/3.0/",
    "credit": "\"Jeff-Walkabout-Denver.mp3\" by kerouacsamerica (Kerouac's America, SUNY Potsdam), Freesound high-quality preview",
    "excerptStartSeconds": 6.6,
    "excerptEndSeconds": 22.8,
    "sampleRate": 44100,
    "channels": 1,
    "excerptSha256": "9d40034e766f539f6542badebf718f7770ec88e8788e3bf839675b4d3d7f3375",
    "channelNote": "The source's right channel is its left channel with inverted polarity (L/R correlation -1.00), so a mono mix cancels the voice. The excerpt is the left channel only (ffmpeg pan=mono|c0=c0); that mono file is what was uploaded and what plays as Original."
   },
   "restoration": {
    "app": "Diffio web app (local stack on the dev Mac mini fleet)",
    "model": "Diffio 4.5 Pro",
    "projectName": "Street walkabout (left channel)",
    "resultSeenAtUtc": "2026-10-19T22:58:28.372Z",
    "secondsFromCreateToResult": 26.8,
    "exports": {
     "speech": {
      "name": "Restored speech",
      "sha256": "51dff251e0b89fcf59304ad11655748f6c6c143eb4405f383104b93fadf5e471"
     },
     "background": {
      "name": "Background noise",
      "sha256": "2dd47ce9b209e02825f6097ca37012a78f4d2a92a5513aaeaa4f0a0fc9cd710a"
     }
    }
   },
   "playback": {
    "format": "MP3 mono 44.1 kHz 96 kb/s",
    "durationSeconds": 16.2,
    "measuredLufsBeforeGain": {
     "original": -24.1,
     "speech": -22.0
    },
    "fixedGainDb": {
     "original": 3.3,
     "speechAndBackground": 2.0
    },
    "quietestFifteenthPercentileDb": {
     "original": -34.4,
     "restoredSpeech": -50.9
    },
    "processing": "Fixed gain only. No compression, limiting or EQ."
   },
   "noise": "Street hiss, passing vehicles, hum and wind on a walking recorder (AudioSet classifier on the exported background: white noise, vehicle, hum, wind)",
   "transcriptCheck": {
    "method": "faster-whisper large-v3 (and large-v3-turbo) on the uploaded excerpt and on the restored speech, fillers (uh, um) ignored",
    "wordMatchRatio": 0.962,
    "differences": [
     "- -> of",
     "this -> the"
    ],
    "note": "The page quotes the restored words."
   },
   "qualityCheck": {
    "method": "ScoreQ no-reference MOS (adapt_nr_telephone, diffio-mac bench scoreq.py and windowing.py: 10 s windows, last window end-anchored, mean) on the uploaded excerpt and the exported restored speech; higher is better. 'Leveled' is the bench's v2 loudness_normalize copy.",
    "original": 3.449,
    "restored": 4.589,
    "restoredMinWindow": 4.538,
    "improvement": 1.14,
    "originalLeveled": 3.471,
    "restoredLeveled": 4.581
   }
  },
  {
   "slug": "station-fans",
   "label": "Air handling and fans",
   "source": {
    "excerptStartSeconds": 17.5,
    "excerptEndSeconds": 33.7,
    "sampleRate": 48000,
    "channels": 2,
    "excerptSha256": "b7cd5b41b2862db8aece5fb4e519b47a9db619f3d45bbcddf4411c06c8b52307"
   },
   "restoration": {
    "app": "Diffio web app (local stack on the dev Mac mini fleet)",
    "model": "Diffio 4.5 Pro",
    "createdAtUtc": "2026-10-19T17:32:48.890Z",
    "secondsFromCreateToResult": 18.3,
    "exports": {
     "speech": {
      "name": "Restored speech",
      "sha256": "0c0c6e02b49c6fa0b9ae7a883009f6a69a09acabed2a3eb2e78e3900b1c975e0"
     },
     "background": {
      "name": "Background noise",
      "sha256": "0b42dc5b91952c1bcc811eda7ff62aac46f34005c907ca8d6e920ace962b035c"
     }
    }
   },
   "playback": {
    "format": "MP3 mono 44.1 kHz 96 kb/s",
    "durationSeconds": 16.2,
    "measuredLufsBeforeGain": {
     "original": -14.0,
     "speech": -22.7
    },
    "fixedGainDb": {
     "original": -6.7,
     "speechAndBackground": 2.0
    },
    "quietestFifteenthPercentileDb": {
     "original": -33.6,
     "restoredSpeech": -45.1
    },
    "processing": "Fixed gain only. No compression, limiting or EQ."
   },
   "qualityCheck": {
    "method": "ScoreQ no-reference MOS (adapt_nr_telephone, diffio-mac bench scoreq.py and windowing.py: 10 s windows, last window end-anchored, mean) on the uploaded excerpt and the exported restored speech; higher is better. 'Leveled' is the bench's v2 loudness_normalize copy.",
    "original": 3.098,
    "restored": 4.583,
    "restoredMinWindow": 4.536,
    "improvement": 1.485,
    "originalLeveled": 3.189,
    "restoredLeveled": 4.598
   },
   "transcriptCheck": {
    "method": "faster-whisper large-v3-turbo on the page's original and restored MP3s",
    "wordMatchRatio": 1.0
   }
  }
 ],
 "walkthrough": {
  "files": [
   "walkthrough.mp4",
   "walkthrough.webm",
   "walkthrough-poster.webp",
   "walkthrough.vtt",
   "walkthrough-transcript.txt"
  ],
  "builtWith": "HyperFrames 0.8.179, 1280x720 master at 30 fps, published as H.264 (CRF 30, mono AAC 80 kb/s) and VP9 (CRF 42, mono Opus 64 kb/s)",
  "durationSeconds": 46.4,
  "storyboard": [
   "0:00 beach-wind original, then the app: upload, restoring, result Original, Restored, Background amount 25%",
   "0:24 storm-diner: 12.85 to 17.15 s of the sample, original then restored",
   "0:33 night-street: 10.40 to 14.40 s of the sample, original then restored",
   "0:41 end card"
  ],
  "soundtrack": "Segments cut from the page samples' original excerpts and Diffio exports with the page's fixed gains; 15 ms fades at each cut. No other processing.",
  "renderedOn": "2026-10-19"
 },
 "credits": "credits.txt",
 "notUsed": [
  {
   "slug": "street-line",
   "source": "https://commons.wikimedia.org/wiki/File:2007-07-15_-_Kwik-E-Mart_Interview_-_William.ogg",
   "reason": "Published until October 19, 2026, then removed: the restored speech dropped the interviewee's replies (word match 0.75)."
  },
  {
   "slug": "train-rumble",
   "source": "https://freesound.org/people/kerouacsamerica/sounds/31551/",
   "reason": "Published until October 19, 2026, then removed: the restored speech scored 3.69 on ScoreQ, well under the kept samples (about 4.5)."
  },
  {
   "slug": "cafe-owner",
   "source": "https://commons.wikimedia.org/wiki/File:Kenyan_Caf%C3%A9_Employs_People_with_Cognitive_Disorders.webm",
   "reason": "Passed (ScoreQ 2.67 to 4.92, word match 0.95); the diner was chosen for chatter, so it was left out."
  },
  {
   "slug": "station-hall",
   "source": "https://freesound.org/people/kerouacsamerica/sounds/31556/",
   "reason": "Passed (ScoreQ 3.49 to 4.78, word match 1.00) but the noise was mild, so it was left out."
  },
  {
   "slug": "field-kitchen",
   "source": "https://commons.wikimedia.org/wiki/File:Staff_Sgt_Christopher_Rivera_55th_Quartermaster_Mobile_Kitchen_Interview_(1000795).webm",
   "reason": "Word match 0.83 (one phrase changed), so it was left out."
  },
  {
   "slug": "market-demo",
   "source": "https://commons.wikimedia.org/wiki/File:20160916-AMS-LSC-9001_USDA_Farmers_Market_and_Ocean_Spray_Cooperative_Cranberry_Bog_Harvest.webm",
   "reason": "Words matched, but the restored speech scored 3.82 on ScoreQ, so it was left out."
  },
  {
   "slug": "flood-rain",
   "source": "https://commons.wikimedia.org/wiki/File:Flooded_Residents_Face_More_Rain_in_Texas.webmhd.webm",
   "reason": "The restored speech changed words (word match 0.81), so it was left out."
  },
  {
   "slug": "drama-class",
   "source": "https://freesound.org/people/12572598/sounds/449511/",
   "reason": "The restored speech changed words (word match 0.71), so it was left out."
  },
  {
   "slug": "phone-memo",
   "source": "https://freesound.org/people/buggly/sounds/790696/",
   "reason": "No background noise to remove (the exported background was silent), so it was left out."
  }
 ]
}