Caller state determination
In a call center, voice recordings of calls with callers can be analyzed to determine an emotional state metric, e.g. based on the pitch, volume, etc. of the voice sample waveform. A sentiment metric can be determined from words spoken in the sample. The emotional state metric and the sentiment metric may be combined to generate a caller state metric. The caller state metric may trigger subsequent actions, such as transferring calls to call agents or supervisors, and identifying points within voice scripts that can be improved.
1. A call center configured to be engaged in a call with a caller, the call center comprising:
a memory storing instructions; and
a processor that executes the instructions to:
analyze a sample of a plurality of voice recordings received from a user over a period of time:
generate, for each sample, an emotional state metric and a sentiment state metric based on the analysis of each sample;
generate, for each sample, a caller state metric based on a combination of the emotional state metric and the sentiment state metric;
identify a change in the caller state metric between each sample; and
identify an action to be taken by the call center based on the identified changes between the each caller metric.
2. The call center of claim 1 , wherein, when the processor is to generate the emotional state metric, the processor further is to:
generate the emotional state metric from a waveform of the sample.
3. The call center of claim 2 , wherein, when the processor is to generate the emotional state metric from a waveform, the processor further is to:
identify at least one of a pitch and a volume of the sample based on the waveform.
4. The call center of claim 1 , wherein, when the processor is to generate the sentiment state metric, the processor further is to:
convert the sample into text;
identify an occurrence of words that are categorized according to a corresponding sentiment; and
generate the sentiment state metric based on a frequency of occurrence of the words.
5. The call center of claim 4 , wherein the words that are categorized according to a corresponding sentiment are stored in a database.
6. The call center of claim 1 , wherein the processor further is to:
divide each voice recording, of the plurality of voice recordings into a plurality of samples; and
generate a plurality of the caller metrics, each corresponding to a sample of the plurality of samples.
7. The call center of claim 6 , wherein the processor further is to:
identify changes in the plurality of caller metrics.
8. A method for processing calls within a call center, the method comprising:
recording, by the call center, a plurality of voice calls from a user over a period of time;
analyzing, by the call center, a sample of each voice call of the plurality of voice calls:
generating, by the call center, and for each sample of the plurality of samples, an emotional state metric and for the a sentiment state metric based on the analyzing;
generating, by the call center, and for each sample, a caller state metric based on a combination of the emotional state metric and the sentiment state metric;
identifying, by the call center, a change in the caller state metric between each sample; and
identifying, by the call center, an action to be taken by the call center based on the identified changes between the each caller metric.
9. The method of claim 8 , wherein the generating the emotional state metric further comprises:
identifying one or more sound parameters of the sample; and
identifying the emotional state based on the one or more sound parameters.
10. The method of claim 9 , wherein the one or more sound parameters comprises at least one of:
a pitch of the sample; and
a volume of the sample.
11. The method of claim 9 , comprising:
identifying a change in the one or more sound parameters occurring during the sample; and
identifying a change in the emotional state metric based on the change.
12. The method of claim 8 , wherein the generating the sentiment state metric further:
converting the sample to text;
identifying an occurrence of words that are categorized according to a corresponding sentiment; and
generating the sentiment state metric based on a frequency of occurrence of the words.
13. The method of claim 12 , wherein the words that are categorized according to a corresponding sentiment are stored in a database.
14. The method of claim 8 , comprising:
dividing each voice call, of the plurality of voice calls, into a plurality of samples; and
generating a plurality of the caller metrics, each corresponding to a sample of the plurality of samples.
15. The method of claim 14 , comprising:
identifying changes in the plurality of caller metrics.
16. The method of claim 15 , comprising:
adjusting a script used by the call center to interact with the user based on the identifying changes in the plurality of caller metrics for the plurality of samples.
17. A non-transitory computer readable medium storing instructions that when executed by a processor of a call center cause the processor to perform:
analyzing a sample of each audio input, of a plurality of audio inputs, received from a user of the call center over a period of time;
generating, for each sample, an emotional state metric and a sentiment state metric;
generating, for each sample, a caller state metric based on a combination of the emotional state metric and the sentiment state metric
identifying a change in the caller state metric between each sample; and
identifying, by the call center, an action to be taken by the call center based on the identified changes between the each caller metric.
18. The non-transitory computer readable medium of claim 17 , wherein the instructions further cause the processor to perform:
identifying a change in the caller state metric across a plurality of samples taken from the audio input.
19. The non-transitory computer readable medium of claim 17 , wherein, when the processor is to perform the generating the sentiment state metric, the instructions further cause the processor to perform:
converting the audio input into text;
identifying an occurrence of words that are categorized according to a corresponding sentiment; and
generating the sentiment state metric based on a frequency of occurrence of the words.
20. The non-transitory computer readable medium of claim 17 , wherein, when the processor is to perform the generating the emotional state metric, the instructions further cause the processor to perform:
identifying a waveform of the audio input; and
identifying the emotional state metric based on at least one of a pitch and a volume identified based on the waveform.