tpro/README_PYPI.md

tpro

Transcript Processing! tpro takes JSON-formatted transcripts produced by various speech-to-text services and converts them to various standardized formats.

Installation and Usage

Non-pip Requirement: Stanford NER JAR

• download and unzip this
• put these files in in /usr/local/bin/:
  • stanford-ner.jar
  • classifiers/english.all.3class.distsim.crf.ser.gz
• you might have to update Java on Linux

Pip

$ pip install tpro

Usage

$ tpro --help

Usage: tpro [OPTIONS] TRANSCRIPT_DATA_PATH OUTPUT_PATH
        [amazon|gentle|speechmatics|google] [universal|vo]

Options:
  -p, --print-output    pretty print the transcript, breaks pipeability
  --language-code TEXT  specify language, defaults to en-US.
  --help                Show this message and exit.

Example

$ cat transcript.json 

  { "job": {
      "lang": "en",
      "user_id": 2152310,
      "name": "recording.mp4",
      "duration": 7,
      "created_at": "Mon Nov 12 14:57:06 2018",
      "id": 9871364
    },
    "speakers": [
      {
        "duration": "6.87",
        "confidence": null,
        "name": "M2",
        "time": "5.98"
      }
    ],
    "words": [
      {
        "duration": "0.13",
        "confidence": "0.670",
        "name": "Hello",
        "time": "5.98"
      },
      {
        "duration": "0.45",
        "confidence": "1.000",
        "name": "there",
        "time": "6.14"
      }
  ]
  
} 

$ tpro transcript.json converted_transcript.json speechmatics universal_transcript

[
    {
        "start": 5.98,
        "end": 6.11,
        "confidence": 0.67,
        "word": "Hello",
        "always_capitalized": false,
        "punc_after": false,
        "punc_before": false
    },
    {
        "start": 6.14,
        "end": 6.59,
        "confidence": 1.0,
        "word": "there",
        "always_capitalized": false,
        "punc_after": false,
        "punc_before": false
    }
]

☝☝☝ There\'s your transcript, which was saved to converted_transcript.json.

STT Services

• Speechmatics
• Amazon Transcribe
• Gentle
• Google Cloud Speech-to-Text

Planned

• Watson
• Mozilla's new open-source STT thing

Output Formats

• Universal Transcript (JSON)
• viraloverlay (JSON)

Planned

• Word (.doc, .docx)
• text files
• SRT (subtitles)
• Draft.js JSON