Skip to content

Latest commit

 

History

History
89 lines (64 loc) · 5.32 KB

freesound.md

File metadata and controls

89 lines (64 loc) · 5.32 KB

Freesound Data Card

Data Collection

Source Collecting Method
Freesound website 1.Scrape from the Freesound website and get the metadata
2. Download according to URLs provided in the metadata file, using a python downloading script

Preprocessing Principles

You may refer to freesonud_preprocess.py for all the details. Here is a concise summary:

For each audio, there are 6 fields in the metadata file, respectively named id, title, tags, description, username, and download_url. We retrieve information from these 6 fields and form a 3-field .json file for each audio. Here are some audio-json pairs selected from the processed dataset:

1.mov 
{
    "text": [
    "DSI Tetra - Sample and Hold Me - B4 (Sample & Hold Me-71-127.", 
    "Single note sampled from an analog synthesizer by Modular Samples."
        ],

    "tag": [
    "multisample", "single-note", "synthesizer", "DSI-Tetra", "midi-note-71", "B4"
    ], 
    "original_data": {
        "tags": ["multisample", "single-note", "synthesizer", "DSI-Tetra", "midi-note-71", "B4"], 
        "description": "Single note sampled from an analog synthesizer by Modular Samples.<br>Modular Samples provides samples of vintage and modern synthesizers for Apple EXS24, Native Instruments Kontakt, Reason and Live samplers, with over 50 gigabytes of public domain content.<br>Sampler files and sound packs are also available at <a href =\"http://modularsamples.com\" target=\"_blank\">http://modularsamples.com</a>.<br><br>Synthesizer: DSI Tetra<br>Patch name (pack): Sample and Hold Me<br>Note: B4<br>Midi note: 71<br>", 
        "username": "modularsamples", 
        "download_url": "https://freesound.org/apiv2/sounds/282776/download/", 
        "title": "DSI Tetra - Sample and Hold Me - B4 (Sample & Hold Me-71-127.wav)", 
        "id": 282776
            }
        }
2.mov 
{
    "text": [
        "futuresoundfx-795.", "Sci-Fi Futuristic Sound Effects From Stolting Media Group."
        ], 
    "tag": [
        "Home-Videos", "DVD", "pod-Cast", "Sound-Effects", "alien-sound-effects", "Remixing", "space", "TV", "media", "Screen", "Video", "Music-Production", "fx", "Recording", "stolting-media-group", "Broadcasting", "effects", "Futuristic", "Alien", "Future", "Radio", "Film"
        ], 
    "original_data": {
        "tags": ["Home-Videos", "DVD", "pod-Cast", "Sound-Effects", "alien-sound-effects", "Remixing", "space", "TV", "media", "Screen", "Video", "Music-Production", "fx", "Recording", "stolting-media-group", "Broadcasting", "effects", "Futuristic", "Alien", "Future", "Radio", "Film"], 
        "description": "Sci-Fi Futuristic Sound Effects From Stolting Media Group. These samples available from <a href=\"http://freesound.org\" rel=\"nofollow\">freesound.org</a> are in mp3 format and for non commercial use only with credit to <a href=\"http://StoltingMediaGroup.com\" rel=\"nofollow\">StoltingMediaGroup.com</a> For a commercial license and high quality WAV format visit <a href=\"http://www.stoltingmediagroup.com\" rel=\"nofollow\">http://www.stoltingmediagroup.com</a>", 
        "username": "stoltingmediagroup", "download_url": "https://freesound.org/apiv2/sounds/158824/download/", 
        "title": "futuresoundfx-795.mp3", 
        "id": 158824
        }
    }

I. Json file generation principles

  • text entry Both title and description may contain candidates to fill in the list of text entry, therefore we have two possible sources of caption:

    1. Take the content of title, remove the potential file extension part of it and replace all the underlines between words with spaces. Let it be the first caption of the audio.
    2. Take the first sentence of the content of description. If this sentence contains HTML tags, then discard it. Otherwise, we will let it be the second caption for the audio

    As a result, the text entry eventually looks like "text": [processed title, first sentence of description without HTML tags]. When training we will randomly choose one of them in the list to be the caption.

  • tag entry Just take tags: field of the metadata file split by delimiter , as the content of tag entry.

  • original_data entry All the six fields’ contents are stored here.

II. Audio filtering principles

  1. Keep samples no longer than 3 minutes, and discard the rest.
  2. Keep audios with sampling rate higher than 16KHZ and discard the rest.
  3. Discard all audios failed to be read by soundfile.read() method or denied by FFmpeg while processing.

III. Audio format specifications

After the preprocessing work, all audio files should be in FLAC format with sampling rate of 48KHZ. (Processed by ffmpeg).