ASR-enhanced speech compression/archiving
A process for compressing an audio speech signal utilizes ASR processing to generate a corresponding text representation and, depending on confidence in the corresponding text representation, selectively applies more, less, or no compression to the audio signal. The result is a compressed audio signal, with corresponding text, that is compact and well suited for searching, analytics, or additional ASR processing.
1. A computer-implemented process for digitally compressing an audio signal that includes human speech utterances, the process comprising at least the following computer-implemented acts:
(1) receiving a digitally encoded audio signal that includes human speech utterances;
(2) identifying portions of the digitally encoded audio signal that correspond to the speech utterances;
(3) performing automatic speech recognition (ASR) processing to produce, for each utterance, a corresponding (i) original audio representation, (ii) text representation, and (iii) ASR confidence that represents a likelihood that the text representation accurately captures all spoken words contained in the corresponding audio representation;
(4) forming a compressed audio signal by concatenating the utterances and, for each utterance, only including either its original audio representation or a compressed version of its original audio representation;
(5) storing the compressed audio signal in a non-transitory, computer-readable medium; and
(6) wherein, for each utterance, determining whether to include its original audio representation or a compressed version of its original audio representation in the compressed audio signal comprises determining whether the utterance's corresponding ASR confidence exceeds a predetermined threshold value.
2. A process, as defined in claim 1 , wherein the ASR confidence values are derived from normalized likelihood scores.
3. A process, as defined in claim 1 , wherein the ASR confidence values are computed using an N-best homogeneity analysis.
4. A process, as defined in claim 1 , wherein the ASR confidence values are computed using an acoustic stability analysis.
5. A process, as defined in claim 1 , wherein the ASR confidence values are computed using a word graph hypothesis density analysis.
6. A process, as defined in claim 1 , wherein the ASR confidence values are derived from associated state, phoneme, or word durations.
7. A process, as defined in claim 1 , wherein the ASR confidence values are derived from language model (LM) scores or LM back-off behaviors.
8. A process, as defined in claim 1 , wherein the ASR confidence values are computed using a posterior probability analysis.
9. A process, as defined in claim 1 , wherein the ASR confidence values are computed using a log-likelihood-ratio analysis.
10. A process, as defined in claim 1 , wherein the ASR confidence values are computed using a neural net that includes word identity and aggregated words as predictors.
11. A process, as defined in claim 1 , wherein determining, for each utterance, whether to include its original audio representation or a compressed version of its original audio representation in the compressed audio signal additionally comprises determining whether a person speaking the utterance can be identified as belonging to group of known persons-of-interest.
12. A process, as defined in claim 1 , wherein determining, for each utterance, whether to include its original audio representation or a compressed version of its original audio representation in the compressed audio signal additionally comprises determining whether a person speaking the utterance can be identified in any earlier processed audio recording(s).
13. A process, as defined in claim 1 , wherein determining, for each utterance, whether to include its original audio representation or a compressed version of its original audio representation in the compressed audio signal additionally comprises determining whether the utterance is spoken with an excited, angry, or frustrated tone.
14. A process, as defined in claim 1 , wherein determining, for each utterance, whether to include its original audio representation or a compressed version of its original audio representation in the compressed audio signal additionally comprises determining whether words or phrases in the utterance indicate a negative sentiment.
15. A process, as defined in claim 1 , wherein determining, for each utterance, whether to include its original audio representation or a compressed version of its original audio representation in the compressed audio signal additionally comprises determining whether the utterance contains any word(s)/phrase(s) from a preselected list of trigger word(s)/phrase(s).
16. A process, as defined in claim 1 , wherein determining, for each utterance, whether to include its original audio representation or a compressed version of its original audio representation in the compressed audio signal additionally comprises determining whether a person speaking the utterance can be identified as a child or an elderly person.
17. A process, as defined in claim 1 , wherein performing ASR processing comprises performing an N-best decoding that produces, for each utterance, a corresponding (i) original audio representation, (ii) N alternative text representations, and (iii) N ASR confidence values that represent respective likelihoods that each of the N text representations accurately captures all spoken words contained in the corresponding audio representation.
18. The process of claim 17 , wherein N is an integer value of at least two and less than ten.