VTT Parse

Parse WebVTT subtitle and transcript files into plain text or structured cues

Overview

The VTT Parse tool parses WebVTT (.vtt) subtitle and transcript files — such as those exported by meeting-recording and video platforms — and returns either the plain speech text or a structured array of cues with timestamps. This makes it easy to feed transcripts into summarization, translation, or analysis steps.

Key Features

• VTT_PARSE
  • Parse a WebVTT file from a URL.
  • Inputs: file_url (URL of the WebVTT file) and output_format (txt or array, default txt).
  • txt returns plain speech text with timestamps and markup removed, one cue per line (result.text).
  • array returns structured cues (result.cues), each with index, optional id, start and end timestamps in HH:MM:SS.mmm format, and the cue text.

Authentication

No authentication required. The tool works directly with file URLs.

Example: Meeting Transcript to Summary

- id: upload_transcript
  name: upload_transcript
  tool: INPUT_FILE
  input:
    - name: description
      value: "Upload the meeting recording transcript (.vtt)"

- id: parse_transcript
  name: parse_transcript
  tool: VTT_PARSE
  input:
    - name: file_url
      value: "{{steps.upload_transcript.result.url}}"
    - name: output_format
      value: txt

- id: summarize_meeting
  name: summarize_meeting
  tool: OPENAI_INVOKE
  config:
    - name: version
      value: gpt-4
  input:
    - name: prompt
      value: |
        Please summarize the following meeting transcript:
        1. Key discussion points
        2. Decisions made
        3. Action items with owners

        ==Transcript==
        {{steps.parse_transcript.result.text}}

Example: Structured Cues with Timestamps

- id: parse_subtitles
  name: parse_subtitles
  tool: VTT_PARSE
  input:
    - name: file_url
      value: "https://example.com/subtitles.vtt"
    - name: output_format
      value: array

Each element of {{steps.parse_subtitles.result.cues}} contains index, start, end, and text, so downstream steps can reference specific moments in the recording.

Notes

• Empty cues are skipped; index is assigned sequentially starting from 1 over the remaining cues.
• In txt mode, formatting markup inside cues is stripped and only the text content is kept.
• If the file cannot be downloaded (non-OK HTTP status), the tool returns a DownloadError; if the content is not valid VTT and no cues can be parsed, it returns a ParseError.