Job examples for real-time transcription

Scheduled jobs can be used to consume video or audio from a source continuously. Two examples are given: one transcribes a radio webstream, and the other creates a media source first and then creates a job to capture the radio stream for audio playback.

Engines used in the examples for real-time transcription

The code example below creates a scheduled job that transcribes a radio stream in near real-time and returns the job ID. You can use the ID later to connect to a WebSocket and retrieve the output.

Engines used

These engines can be used in a flow for real-time transcription:

Example of real-time transcription job

The code example below creates a scheduled job that transcribes a radio stream in near real-time and returns the job ID. You can use the ID later to connect to a WebSocket and retrieve the output.

Variables to declare for createScheduledJob mutation

{
 "name": "Transcription Schedule Job",
 "clusterId": "rt-9d7a5d1b-ffe0-4d71-a982-190522cdf273", // Cluster ID from aiWARE Environment
 "url": "https://stream.revma.ihrhls.com/zc389", // Stream URL
 "startDateTime": "2022-01-20T22:20:00.000Z" // Time in UTC that you want the job to start.
}

GraphQL mutation to create the scheduled job

    mutation CreateScheduleJobFromEngine(
      $clusterId: ID!
      $url: String!
      $name: String!
      $startDateTime: DateTime!
    ) {
      createScheduledJob(
        input: {
          name: $name
          runMode: Recurring
          weeklyScheduleParts: [ 
            { scheduledDay: Monday, startTime: "04:00", stopTime: "17:00" }
            { scheduledDay: Tuesday, startTime: "04:00", stopTime: "17:00" }
            { scheduledDay: Wednesday, startTime: "04:00", stopTime: "17:00" }
            { scheduledDay: Thursday, startTime: "04:00", stopTime: "17:00" }
            { scheduledDay: Friday, startTime: "04:00", stopTime: "17:00" }
          ]
          jobTemplates: {
            clusterId: $clusterId
            jobConfig :{
              maxTDODuration: 5
            }
            taskTemplates: [
              {
                # TVR Adapter: stream URL contents in 32K chunks
                engineId: "74dfd76b-472a-48f0-8395-c7e01dd7f255"
                payload: {
                  url: $url
                  isLive: true
                  notUseCurl: true
                  chunkSize: 8000
                  allowOvershot: true
                }
                ioFolders: [{ referenceId: "tvrOut", mode: stream, type: output }]
              }
              {
                # Engine: run AI on stream
                engineId: "09f04ded-591d-4ef3-b45f-eb93b79c7c39"
                payload: { 
                  includePartial: false, 
                  outputInRealTime: false
                }
                ioFolders: [
                  { referenceId: "engineIn", mode: stream, type: input }
                  { referenceId: "engineOut", mode: chunk, type: output }
                ]
              }
              {
                # Output Writer: aggregate chunks and store in TDO
                engineId: "8eccf9cc-6b6d-4d7d-8cb3-7ebf4950c5f3"
                executionPreferences: { parentCompleteBeforeStarting: true }
                ioFolders: [{ referenceId: "owIn", mode: chunk, type: input }]
              }
            ]
            routes: [
              {
                parentIoFolderReferenceId: "tvrOut"
                childIoFolderReferenceId: "engineIn"
              }
              {
                parentIoFolderReferenceId: "engineOut"
                childIoFolderReferenceId: "owIn"
              }
            ]
          }
          startDateTime: $startDateTime
        }
      ) {
        id
      }
    }

The response looks like this: 

    {
     data: {
       createScheduledJob: {
         id: 1
       }
     }
    }

Example of an audio capture job

This example creates a scheduled job for audio capture from a radio stream and then returns the job ID. You can use the ID later to connect to a WebSocket and hear the audio. To do this:

1. Create a media source with a stream URL.
2. Input the URL into the audio job payload.
3. Create an audio job.

Engines used

These engines can be used to capture real-time audio for playback:

Variables to declare for createMediaSource mutation

  {
   "name": "Stream URL Example 1", // Name of the media source.
   "url": "https://stream.revma.ihrhls.com/zc389", // Stream URL
   "timezone": "America/Los_Angeles" // The timezone that you want to use that will correlate with the scheduled job you create.
  }

GraphQL mutation to create the media source

    mutation CreateMediaSource($name: String!, $url: String!, $timezone: String!) {
     createSource(
       input: {
         name: $name
         sourceTypeId: "1" # Radio Stream Source Type
         details: { liveTimezone: $timezone, radioStreamUrl: $url }
       }
     ) {
       id
     }
    }

The response looks like this: 

    {
     data: {
       createSource: {
         id: 1
       }
     }
    }

Variables to declare for scheduled job mutation

    {
     "name": "Transcription Schedule Job",
     "clusterId": "rt-9d7a5d1b-ffe0-4d71-a982-190522cdf273", // Cluster ID from aiWARE Environment
     "url": "https://stream.revma.ihrhls.com/zc389", // Stream URL
     "sourceIdJob": 1, // The media source id
     "sourceIdPayload": "1", // The media source id as a string
     "startDateTime": "2022-01-20T22:20:00.000Z" // Time in UTC that you want the job to start.
    }

GraphQL mutation to create the scheduled job

    mutation CreateScheduleJobFromEngine(
     $clusterId: ID!
     $name: String!
     $sourceIdJob: ID
     $sourceIdPayload: String
     $startDateTime: DateTime!
    ) {
     createScheduledJob(
       input: {
         name: $name
         runMode: Recurring
         weeklyScheduleParts: [
           { scheduledDay: Monday, startTime: "04:00", stopTime: "17:00" }
           { scheduledDay: Tuesday, startTime: "04:00", stopTime: "17:00" }
           { scheduledDay: Wednesday, startTime: "04:00", stopTime: "17:00" }
           { scheduledDay: Thursday, startTime: "04:00", stopTime: "17:00" }
           { scheduledDay: Friday, startTime: "04:00", stopTime: "17:00" }
         ]
         jobTemplates: {
           clusterId: $clusterId
           jobConfig: {
             maxTDODuration: 5
             sourceData: { sourceId: $sourceIdJob }
           }
           taskTemplates: [
             {
               # TVR Adapter: stream URL contents in 32K chunks
               engineId: "74dfd76b-472a-48f0-8395-c7e01dd7f255"
               payload: {
                 sourceId: $sourceIdPayload
                 isLive: true
                 notUseCurl: true
                 allowOvershot: true
               }
               ioFolders: [{ referenceId: "tvrOut", mode: stream, type: output }]
             }
             {
               # Stream Ingestor 2 (SI2) Playback engine to store playback segments
               engineId: "352556c7-de07-4d55-b33f-74b1cf237f25"
               ioFolders: [
                 { referenceId: "playbackInput", mode: stream, type: input }
               ]
               executionPreferences: { parentCompleteBeforeStarting: true }
             }
           ]
           routes: [
             {
               parentIoFolderReferenceId: "tvrOut"
               childIoFolderReferenceId: "playbackInput"
             }
           ]
         }
         startDateTime: $startDateTime
       }
     ) {
       id
     }
    }

The response looks like this: 

    {
     data: {
       createScheduledJob: {
         id: 2
       }
     }
    }

What to do next

Connect to a WebSocket

• Engines used in the examples for real-time transcription
• Example of real-time transcription job
• Example of an audio capture job