Voice Features

This optional feature returns a set of per-word metrics, used in data science to serve as input attributes to a predictive model. Voice Features is supported for all regions of English as well as US and Latin American Spanish.

Uses for Voice Features

Voice Features metrics may be used in multiple ways. One example would be determining that a caller raised their voice and spoke at a higher pitch during a call, suggesting they may be upset. Making this determination requires that you process these per-word metrics to: 1) Create a function that transforms per-word metrics into utterance or time-leveled metrics 2) Gather baseline values for the caller 3) Track changes over time relative to the baseline 4) Set thresholds, and assign meaning to them.

Enabling Voice Features

Enable Voice Features by including "voiceFeatures" in the features array under speechModel when making a call to POST /media.

{
  "speechModel": {
    "features": [
        "voiceFeatures"
    ]
  }
}

The configuration contains the following fields:

  • speechModel: speech model section
  • features: speech model features to enable

Voice Features Results

When you GET the media results, the response will include the “frq” array, populated with “f”(frequency),“e”(energy), and “v”(volume) values for each word.

“f” is the frequency in Hertz,computed from the dominant and secondary peaks of each word. The maximum value is 8khz or 16khz, depending on submitted audio.

“e” is the relative energy (amplitude) of the frequency. The value will be between 0 and 1.

“v” is the relative volume. Volume is determined by a formula where v = A * M. A is the average between the words timing and the frequencies amplitude and M is the maximum amplitude value from any frequency and spectrum frames within the words. The value can be unlimited but it is typically between 0 and 2. Any value greater than 2 can be cropped to 2.

For example:

{
  "mediaId": "bc14632d-e81b-4673-992d-5c5fb6573fb8",
  "status": "finished",
  "dateCreated": "2017-06-22T19:18:49Z",
  "dateFinished": "2017-06-22T19:19:27Z",
  "mediaContentType": "audio/x-wav",
  "length": 10031,
  "transcript": {
     "words":[
        {
            "p": 0,
            "s": 1880,
            "c": 0.461,
            "frq": [
              {
                "e": 1,
                "f": 260.942
              },
              {
                "e": 0.347,
                "f": 521.807
              }
            ],
            "e": 2180,
            "v": 14.876,
            "w": "Because"
          }
     ]
  }
}