System and Method of Text Zoning
A method of zoning a transcription of audio data includes separating the transcription of audio data into a plurality of utterances. A that each word in an utterances is a meaning unit boundary is calculated. The utterance is split into two new utterances at a work with a maximum calculated probability. At least one of the two new utterances that is shorter than a maximum utterance threshold is identified as a meaning unit.
1 . A method of zoning a transcription of audio data, the method comprising:
separating the transcription of audio data into a plurality of utterances;
identifying utterances of the plurality of utterances that are shorter than a predetermined minimum threshold as meaning units;
calculating a probability that each word in an utterance of the plurality of utterances which is longer than the predetermined minimum threshold is a meaning unit boundary;
splitting the utterance longer than the predetermined minimum threshold into two new utterances at a word with a maximum calculated probability; and
identifying at least one of the two utterances that is shorter than a maximum utterance threshold as a meaning unit.
2 . The method of claim 1 , wherein calculating the probability that each word in the utterance longer than the predetermined minimum threshold is a meaning unit boundary is further based upon at least a linguistic model.
3 . The method of claim 2 , wherein the linguistic model comprises statistics, distributions, or frequencies of word pairs or word triplets.
4 . The method of claim 2 , wherein the linguistic model comprises probability of words to form the beginning or end of a meaning unit.
5 . The method of claim 2 wherein calculating the probability that each word in the utterance longer than the predetermined minimum threshold is a meaning unit boundary is further based upon an acoustic model.
6 . The method of claim 2 , further comprising receiving audio data and decoding the audio data to create the transcription of audio data.
7 . The method of claim 5 , wherein at least the linguistic model is used when decoding the audio data to create the transcription of audio data.
8 . The method of claim 1 , further comprising applying speech analytics to the meaning unit to identify at least one of context or content of the meaning unit.
9 . The method of claim 1 , further comprising applying speech analytics to identified meaning units to group the meaning units into call segments.
10 . The method of claim 9 , further comprising applying speech analytics to the identified meaning units to identify dialog acts within the identified meaning units.
11 . The method of claim 1 , wherein the predetermined minimum threshold is thirty words.
12 . The method of claim 1 , further comprising:
selecting utterances of the plurality that are longer than the predetermined minimum threshold for subdivision; and
splitting the selected utterances of the plurality into widows, each window being twice the maximum utterance threshold.
13 . The method of claim 12 , wherein calculating the probability that each word in an utterance longer than the predetermined minimum threshold is a meaning unit boundary is calculated for each word in each window.
14 . The method of claim 13 , further comprising applying at least one of a linguistic exception and an acoustic exception to the two new utterances.
15 . The method of claim 14 , wherein the at least one linguistic exception comprises a minimum meaning unit boundary probability or a minimum meaning unit boundary probability differential.
16 . The method of claim 14 , wherein the at least one acoustic exception comprises an identification of a pause between adjacent utterances in the transcription of the audio data.
17 . A method of zoning, a transcription of audio data, the method comprising:
separating the transcription of audio data into a plurality of utterances
identifying utterances of the plurality of utterances that are shorter than a predetermined minimum threshold as meaning units;
selecting utterances of the plurality of utterances that are longer than the predetermined minimum threshold for subdivision;
splitting the selected utterances into widows, each window being twice a maximum utterance threshold;
calculating a probability that each word in the plurality of windows is a meaning unit boundary based upon at least a linguistic model applied to each of the plurality of windows;
splitting the selected utterances which are longer than the predetermined minimum threshold into two new utterances at a word with a maximum calculated probability; and
identifying at least one of the two new utterances that is shorter than a maximum utterance threshold as a meaning unit.
18 . The method of claim 17 , wherein the linguistic model comprises probability of words to form the beginning or end of a meaning unit.
19 . The method of claim 18 , further comprising receiving audio data and decoding the audio data with at least the linguistic model to create the transcription of audio data.
20 . The method of claim 19 , further comprising applying at least one of a linguistic exception and an acoustic exception to the two new utterances.