Brief of the project

Work in progress

A module to align stt transcription with accurate text that has got speaker labels.

For when you already have a transcription, eg in a csv file, with speaker names but no timecodes. And you want to add timecodes but not lose the speaker labels (or the accurate text).

Setup

git clone git@github.com:pietrop/align-diarized-text.git

cd align-diarized-text

See docs notes 'analyses' for more info on getting the media and transcripts, in the right format for this.

Usage

npm install align-diarized-text

see example usage in /src/add-timecodes-to-quotes/

const alignDiraizedText = require('./index.js');

const linesWithSpeaker = require('../../sample-data/input-example.json');
const sttJson = require('../../sample-data/stt-transcript.json')

const res = alignDiraizedText(linesWithSpeaker, sttJson);
// do something with output json 

To troubleshoot the aligement you can also use generateInteractiveTranscript, see example usage in /src/generate-html-to-check-alignement/ to generate an index.html file.

input:

1. Json

An Array list of objects with text and speaker attribtues.

[
  {
    "id": "6af9762b-d0aa-42d2-9d6d-1c114f9219db",
    "text": "Thank you. It's good to be here.",
    "speaker": "Elizabeth Warren"
  },
  {
    "id": "b02da04e-044e-436a-8862-166139568136",
    "text": "So I think of it this way, who is this economy really working for? It's doing great for a thinner and thinner slice at the top. It's doing great for giant drug companies. This is not doing great for people are trying to get a prescription filled. It's doing great for people who want to invest in private prisons just not for the African-Americans and Latinx whose families are torn apart whose lives are destroyed and whose communities are ruined.",
    "speaker": "Elizabeth Warren"
  },
...

The text is human accurate transcription, timecodes are missing, and speaker diraziation info is present.

This could be initially originated from a .tsv or .csv file, and converted to json with 'convert-csv-to-json'.

2. STT

STT array of timecoded words, This is generated from video/audio file of the debate. see sample-data folder more.

Output

View sample-data folder for example output.

Something like this

[
  {
    "start": 144.94,
    "end": 168.31,
    "text": "Thank you. It's good to be here.",
    "words": [
      {
        "end": 145.18,
        "start": 144.94,
        "text": "Thank"
      },
      {
        "start": 145.2,
        "end": 145.66,
        "text": "you."
      },
      ...
    ],
    "id": "6af9762b-d0aa-42d2-9d6d-1c114f9219db",
    "speaker": "Elizabeth Warren"
  },
  ...

System Architecture

TBC

Documentation

There's a docs folder in this repository.

docs/notes contains dev draft notes on various aspects of the project. This would generally be converted either into ADRs or guides when ready.

Development env

• npm > 6.1.0
• Node 10 - dubnium

Node version is set in node version manager .nvmrc

Build

NA

Tests

NA

Deployment

NA