Keywords and Topics

VoiceBase can discover the keywords, key phrases, and topics in your recording using a processing known as semantic indexing. The results are known as "semantic knowledge discovery" and it includes keywords and topics.

VoiceBase also supports keyword and key phrase spotting. You can define groups of keywords or key phrases, which are flagged when they are spotted in the recording. For more details, see Keyword Spotting.

Semantic Keywords and Topics

Semantic keywords and topics are discovered automatically (for languages where the feature is supported) and returned with the analytics under the section "knowledge" For example, the knowledge.keywords section may contain an entry like the following:

{
  "keyword": "subscription",
  "relevance": 0.17,
  "mentions" : [
     {
       "speakerName" : "caller",
       "occurrences" : [
          { "s": 34480, "exact": "subscription" },
          { "s": 57340, "exact": "membership" }
       ]
     },
     {
       "speakerName" : "agent",
       "occurrences" : [
          { "s": 40010, "exact": "subscription" }
       ]
     }

  ]
}

In this example, the keyword subscription was spoken by the caller around 34 and 57 seconds into the recording, and by the agent around 40 seconds into the recording.

Topics are reported by grouping a set of keywords that relate to each other. For example, the section knowledge.topics may contain an entry like this:

{
  "topicName": "Solar energy",
  "subTopics": [ ],
  "relevance": 0.004,
  "keywords": [
    {
      "keyword": "solar power",
      "relevance": 0.17,
      "mentions" : [
         {
           "speakerName" : "caller",
           "occurrences" : [
              { "s": 34480, "exact": "solar power" },
              { "s": 57340, "exact": "solar energy" }
           ]
         },
         {
           "speakerName" : "agent",
           "occurrences" : [
              { "s": 40010, "exact": "solar power" }
           ]
         }

      ]
    },
    {
      "keyword": "solar water heating",
      "relevance": 0.17,
      "mentions" : [
         {
           "speakerName" : "caller",
           "occurrences" : [
              { "s": 134480, "exact": "solar water heater" },
              { "s": 157340, "exact": "solar thermal collector" }
           ]
         }
      ]
    }
  ]
}

Relevance

Semantic keywords and topics are scored for relevance. The Keyword relevance score ranges from 0.0 to 1.0, with higher scores indicating higher relevance. The Topic relevance score is an average of the scores of its keywords, and may be any real number.

Keyword relevance is computed by matching detected keywords and key phrases to an extensive taxonomy of potential keywords (VoiceBase's semantic index). The relevance algorithm generally produces higher scores for more frequent, closely matching, and distinctive keywords, while producing lower scores for very common keywords found in the recording.

Enabling Semantic Keywords and Topics

Enable semantic keywords and topic extraction by adding keywords and topics to your media POST configuration.

The configuration attachment should contain the key:

knowledge : Settings for Knowledge Discovery
- enableDiscovery : Switch for enabling/disabling knowledge discovery. The default is false.
- enableExternalDataSources : Switch for allowing the search on sources external to VoiceBase. Users concerned about data privacy or PCI requirements can turn this off. Default is true.

For example:

{  
  "knowledge": {
    "enableDiscovery": true,
    "enableExternalDataSources" : true
  }
}

Examples

Example: Enabling semantic knowledge discovery

The following is an example of posting a media document with semantic keywords and topics extraction enabled.

curl https://apis.voicebase.com/v3/media \
  --header "Authorization: Bearer $TOKEN" \
  --form [email protected] \
  --form configuration='{
    "knowledge": {
      "enableDiscovery": true,
      "enableExternalDataSources" : false
     }
  }'