Handle audio ml:Fields using sc:AudioObject

[marcenacp]

Idea: We want to handle audio ml:Fields using https://schema.org/AudioObject.

Issue 1

Does sc:AudioObject contain all the needed properties? In particular, is bitrate being an sc:Text an issue?

Issue 2

The other challenge is to have ml:Field inherit all the properties of sc:AudioObject.

Solution 1: Add a field audioProperties with all the audio properties.

{
  "@type": "ml:Field",
  "name": "audio",
  ...
  "audioProperties": {
    "@type": "sc:AudioObject",
    "bitrate": "24kbps",
    "encodingFormat": "audio/mpeg"
  }
}

Solution 2: Use additionalProperty.

{
  "@type": "ml:Field",
  "name": "audio",
  ...
  "additionalProperty": [
    {
      "@type": "sc:PropertyValue",
      "name": "bitrate",
      "value": "24kbps"
    },
    {
      "@type": "sc:PropertyValue",
      "name": "encodingFormat",
      "value": "audio/mpeg"
    }
  ]
}

This discussion is related to the issue #240. Cc: @benjelloun @monke6942021 @fineguy.

[fineguy]

I think having bitrate as text field is fine, because it usually includes measure units, e.g. "320Kbps" (common for MP3 files) or "10Mpbs" (common for Full HD videos).

But I'm not sure if sc:AudioObject describes audio data well enough for data processing tasks. For example I don't see anything related to sampling rate or number of channels, which are commonly necessary for correct audio file reading, e.g. in librosa.load or pydub.AudioSegment.from_file.

[marcenacp]

@fineguy Thanks for the answer.

@benjelloun sc:AudioObject doesn't have all the required properties for data processing. What is the best practice? Should we create ml:AudioObject or extend sc:AudioObject with the additional properties (numChan, dtype)?

[benjelloun]

2 cents: Solution 2 seems more in line with schema.org and is more widely applicable, so I would favor that. We should rely on propertyID to identify the property we are talking about, e.g.: "propertyID": "sc:bitrate"

For properties that are not well described by sc:AudioObject, solution 2 can also be used, provided we find other vocabularies that define the right properties. propertyID can refer to the corresponding URLs.

(See examples 11 and 12 in the schema.org/PropertyValue documentation.)