System and method of automated model adaptation
Methods, systems, and computer readable media for automated transcription model adaptation includes obtaining audio data from a plurality of audio files. The audio data is transcribed to produce at least one audio file transcription which represents a plurality of transcription alternatives for each audio file. Speech analytics are applied to each audio file transcription. A best transcription is selected from the plurality of transcription alternatives for each audio file. Statistics from the selected best transcription are calculated. An adapted model is created from the calculated statistics.
1. A method of automated transcription model adaptation, the method comprising:
obtaining audio data from a plurality of audio files;
pre-processing the audio data before transcription to segment the audio data into a plurality of segmented utterances, wherein the segmented utterances are segments of audio data that are likely to be speech separated by segments of audio data that are likely to be non-speech;
applying at least one applied model to the segmented utterances, creating an audio file transcription of the audio data for each of the audio files, wherein each audio file transcription represents a one-best output transcription of each audio file based on the application of the at least one applied model, wherein the one-best output transcription is the most likely transcription based on the applied model;
calculating statistics from each audio file transcription; and
creating an adapted model from the calculated statistics.
2. The method of claim 1 , wherein transcribing the audio data is performed using a large vocabulary continuous speech recognition system.
3. The method of claim 1 , further comprising diarizing at least one audio file transcription to identify at least a first speaker and a second speaker in the at least one audio file transcription, wherein the statistics are calculated from the diarized at least one audio file transcription for the first speaker and the second speaker.
4. The method of claim 3 , wherein the adapted model is a model for the first speaker.
5. The method of claim 1 , wherein the adapted model is an acoustic model.
6. The method of claim 1 , wherein the adapted model is a linguistic model.
7. The method of claim 1 , wherein the applied model is a compiled model.
8. The method of claim 1 , wherein the applied model is a generalized model.
9. The method of claim 1 , wherein the applied model is a specialized model.
10. The method of claim 1 , wherein the one-best output transcription is determined using a Viterbi algorithm.
11. The method of claim 1 , further comprising filtering the audio data prior to calculating statistics.
12. The method of claim 11 , further comprising diarizing the audio data after filtering to identify at least a first speaker and a second speaker in the audio data.
13. The method of claim 11 , wherein the audio data is filtered after transcribing the audio data.
14. The method of claim 1 , further comprising diarizing the audio data to identify at least a first speaker and a second speaker in the audio data.
15. The method of claim 1 , further comprising filtering the at least one audio file transcription to select each audio file transcription for a particular customer service agent.
16. The method of claim 15 , wherein the adapted model is created using audio file transcription for only the audio file transcripts for the particular customer service agent.
17. The method of claim 1 , wherein the method is repeated iteratively.
18. The method of claim 17 , wherein during a first iteration the applied model is a generalized model and during any subsequent iterations the applied model is the adapted model created in the previous iteration.
19. A system for automated adaptation of a transcription model, the system comprising:
an audio data source upon which a plurality of audio data files are stored;
a processor that receives the plurality of audio data files, segments the audio data files into segmented utterances, wherein the segmented utterances are segments of audio data that are likely to be speech separated by segments of audio data that are likely to be non-speech, applies at least one applied model to the segmented utterances, creates an audio file transcription of each of the plurality of audio data files to produce an audio data file transcription, wherein each audio data file transcription represents a one-best output transcription of each audio file based on the application of the at least one applied model, wherein the one-best output transcription is the most likely transcription based on the transcription model; and
a non-transitory computer readable medium communicatively connected to the processor and programmed with computer readable code that when executed by the processor causes the processor to:
calculate statistics from the each audio data file transcription; and
create an adapted model based upon the calculated statistics.
20. A non-transitory computer readable medium programmed with computer readable code that upon execution by a computer processor causes the computer processor to:
obtain audio data from a plurality of audio files;
pre-process the audio data before transcription to segment the audio data into a plurality of segmented utterances, wherein the segmented utterances are segments of audio data that are likely to be speech separated by segments of audio data that are likely to be non-speech;
create an audio transcription of the audio data for each of the audio files, wherein each audio file transcription represents a one-best output transcription of each audio file based on the application of the at least one applied model, wherein the one-best output transcription is the most likely transcription based on the applied model;
calculate statistics from each audio file transcription; and
create an adapted model from the calculated statistics.