Added transcription helpers for extracting text from a canvas

[stephenwf]

Transcription helper.

Will find the following transcriptions:

• VTT as rendering on canvas
• Embedded Annotation page
• External Annotation page
• ALTO annotations (FUTURE)

Cookbook:

• https://iiif.io/api/cookbook/recipe/0017-transcription-av/
• https://iiif.io/api/cookbook/recipe/0219-using-caption-file/
• https://iiif.io/api/cookbook/recipe/0046-rendering/
• https://iiif.io/api/cookbook/recipe/0231-transcript-meta-recipe/
• https://iiif.io/api/cookbook/recipe/0068-newspaper/

Plaintext rendering on canvas:

"rendering": [
  {
    "id": "https://fixtures.iiif.io/video/indiana/volleyball/volleyball.txt",
    "type": "Text",
    "label": {
      "en": [
        "Transcript"
      ]
    },
    "format": "text/plain"
  }
]

VTT annotation body on AV canvases:

"annotations": [
  {
    "id": "https://iiif.io/api/cookbook/recipe/0219-using-caption-file/canvas/page2",
    "type": "AnnotationPage",
    "items": [
      {
        "id": "https://iiif.io/api/cookbook/recipe/0219-using-caption-file/canvas/page2/a1",
        "type": "Annotation",
        "motivation": "supplementing",
        "body": {
          "id": "https://fixtures.iiif.io/video/indiana/lunchroom_manners/lunchroom_manners.vtt",
          "type": "Text",
          "format": "text/vtt",
          "label": {
            "en": [
              "Captions in WebVTT format"
            ]
          },
          "language": "en"
        },
        "target": "https://iiif.io/api/cookbook/recipe/0219-using-caption-file/canvas"
      }
    ]
  }
]

OCR annotations:

• a motivation of supplementing,
• the URI of the OCR file in the id property of the Annotation body, and
• the target set to the applicable Canvas.

{
  "id": "https://iiif.io/api/cookbook/recipe/0068-newspaper/newspaper_issue_1-anno_p1.json-1",
  "type": "Annotation",
  "motivation": "supplementing",
  "body": {
    "type": "TextualBody",
    "format": "text/plain",
    "language": "de",
    "value": "I. 54. Jahrgang"
  },
  "target": {
    "type": "SpecificResource",
    "source": {
      "id": "https://iiif.io/api/cookbook/recipe/0068-newspaper/canvas/p1",
      "type": "Canvas",
      "partOf": [
        {
          "id": "https://iiif.io/api/cookbook/recipe/0068-newspaper/newspaper_issue_1-manifest.json",
          "type": "Manifest"
        }
      ]
    },
    "selector": {
      "type": "FragmentSelector",
      "conformsTo": "http://www.w3.org/TR/media-frags/",
      "value": "xywh=0,376,399,53"
    }
  }
}

OR
Linking Directly to an ALTO File. (FUTURE, NOT IMPLEMENTED)

"rendering": [
  {
    "id": "https://iiif.io/api/cookbook/recipe/0068-newspaper/newspaper_issue_1-alto_p2.xml",
    "type": "Text",
    "format": "application/xml",
    "profile": "http://www.loc.gov/standards/alto/",
    "label": {
      "en": [
        "ALTO XML"
      ]
    }
  }
],

It will produce a standard format for both temporal and plaintext/positional plaintext, including selectors.

interface Transcription {
  id: string;
  source: any;
  plaintext: string;
  segments: Array<{
    text: string;
    textRaw: string;
    granularity?: 'word' | 'line' | 'paragraph' | 'block' | 'page';
    language?: string;
    selector?: ParsedSelector;
    startRaw?: string;
    endRaw?: string;
  }>;
}

ParsedSelector include spatial and temporal information. Either from an annotation or from VTT (very simple parsing at the moment - external libraries for it are heavy). If there is just plaintext by itself, then there are no segments.

A viewer could start with just showing the plaintext, and then implement optional segments later.

Some new helpers too:

• canvasHasTranscriptionSync() - checks if there is a transcription on a canvas without making any network requests
• canvasLoadExternalAnnotationPages() loads and waits for external Annotation Pages
• annotationPageToTranscription() - actual code for fetching the transcription - will also fetch all annotation pages. Recommended to use with Vault (to avoid multiple requests).

[codesandbox-ci]

This pull request is automatically built and testable in CodeSandbox.

To see build info of the built libraries, click here or the icon next to each commit SHA.

[stephenwf]

At the moment, we are losing track of the Annotation target when parsing. It will very likely be the Canvas, but it could be

• Canvas ID
• Media id (complex timeline)
• Choice ID (indicating it works with all choices)

And clients might need to check when they are providing navigation using the selector that it's got the right target.

[stephenwf]

Also need to pass in a language, so that the transcription can check for choices structured like this:
https://iiif.io/api/cookbook/recipe/0074-multiple-language-captions/

[stephenwf]

This still needs more testing, will leave open.