> For the complete documentation index, see [llms.txt](https://autoedit.gitbook.io/documentation/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://autoedit.gitbook.io/documentation/master-1.0.13-1/transcriptions/transcriptions-intro/transcriptions.md).

# Transcription json

## Component/part description

In this section we are looking at the Transcription JSON schema used in autoEdit.

## Related projects.

Connected to this is defining a schema for transcription, to make sure all components that work with this have a defined interface/specification. Eg [BBC Transcript model](https://github.com/bbc/transcript-model)

### Gentle - json

[See appendix for json](/documentation/master-1.0.13-1/appendix-data-structures/gentle-json-transcription-specs.md) example.

### IBM - Json

[See appendix for json](/documentation/master-1.0.13-1/appendix-data-structures/ibm-watson-json-specs.md) example. As well as their [stt api reference](https://www.ibm.com/watson/developercloud/speech-to-text/api/v1/) and documentation.

### Pocketsphinx  - plain text

[See appendix for example](/documentation/master-1.0.13-1/appendix-data-structures/pocketsphinx-results.md) pocketsphinx plain text.

## Implementations Options considered

### Other

An array of words object, to represent lines this could also be a nested array of word objects.

Where the word object at a minimum as a start, end time and text attribute.

### BBC Transcription Model

considered the [bbc/transcript-model](https://github.com/bbc/transcript-model) to make use of [bbc/transcript-editor](https://github.com/bbc/transcript-editor) React component, built on top of Draft.js to improve the word correction feature. 

However  the blocker for now is that changing the data structure for transcription at this stage would involve a pretty substantial refactor. 

{% embed url="<https://trello.com/c/w3MH2HUi>" %}

{% embed url="<https://trello.com/c/3f35GNS2>" %}

## Current implementation

### Transcription domain

autoEdit JSON Transcription schema at a high level it models the objects present in a transcription.

![Transcription modelling diagram](https://440779116-files.gitbook.io/~/files/v0/b/gitbook-legacy-files/o/assets%2F-LKEQtUEJT5OEtyEbMfm%2F-LKESXcFA1aAsli3q3WF%2F-LKES_wU1YbJT0EW_iOH%2FTranscription%20modelling.png?generation=1534636483127889\&alt=media)

In this representation:

* Transcription&#x20;
  * Paragraphs  ← speaker&#x20;
    * Lines&#x20;
      * Words&#x20;

Speakers are associated to paragraphs. Paragraphs are treated as sections of lines.

A list of speakers can also be kept separate, similarly to how IBM Watson stt API returns the results of speaker diarization.

[See Appendix for autoEdit json schema example](/documentation/master-1.0.13-1/appendix-data-structures/autoedit-transcription-json.md).

## What needs refactoring

Name of `paragraph` and `line` attribute are ambiguos. It should be `lines` and `words` instead(?).

### replace array with hash

This is a bigger refactoring but instead of array data structure, could use hash/dictionary.

This way id is the key. and can make use of `key` `value` methods available in js. Lookup speed would improve(?) and could easily get array of values using js method.&#x20;