Automatic insertion of non-verbalized punctuation
First Claim
Patent Images
1. A method of recognizing punctuation in computer-implemented speech recognition in a computer having a processor responsive to instruction for performing the method of recognizing punctuation, the method comprising:
- receiving the instructions by the processor, and executing the instructions for;
performing speech recognition on an utterance to produce a recognition result for the utterance;
identifying a non-verbalized punctuation mark in a recognition result including predicting the non-verbalized punctuation mark using at least one text feature and at least one acoustic feature related to the utterance;
inserting the non-verbalized punctuation mark into the recognition result; and
formatting the recognition result based on the identification of the non-verbalized punctuation mark after the non-verbalized punctuation mark has been inserted in the recognition result;
wherein the acoustic feature includes one or more of a length of a period of silence and a function of pitch of words near the period of silence, the acoustic feature including an average pitch of words near the period of silence and a function of a pitch of words adjacent to the word gap, the acoustic features based on word adjacent to the word gap including the average pitch of the words two back from the word gap and the a ratio of the average pitch of words one forward and one back from the word gap.
5 Assignments
0 Petitions
Accused Products
Abstract
Recognizing punctuation in computer-implemented speech recognition includes performing speech recognition on an utterance to produce a recognition result for the utterance. A non-verbalized punctuation mark is identified in a recognition result and the recognition result is formatted based on the identification.
52 Citations
14 Claims
-
1. A method of recognizing punctuation in computer-implemented speech recognition in a computer having a processor responsive to instruction for performing the method of recognizing punctuation, the method comprising:
-
receiving the instructions by the processor, and executing the instructions for; performing speech recognition on an utterance to produce a recognition result for the utterance; identifying a non-verbalized punctuation mark in a recognition result including predicting the non-verbalized punctuation mark using at least one text feature and at least one acoustic feature related to the utterance; inserting the non-verbalized punctuation mark into the recognition result; and formatting the recognition result based on the identification of the non-verbalized punctuation mark after the non-verbalized punctuation mark has been inserted in the recognition result; wherein the acoustic feature includes one or more of a length of a period of silence and a function of pitch of words near the period of silence, the acoustic feature including an average pitch of words near the period of silence and a function of a pitch of words adjacent to the word gap, the acoustic features based on word adjacent to the word gap including the average pitch of the words two back from the word gap and the a ratio of the average pitch of words one forward and one back from the word gap. - View Dependent Claims (2, 3, 4, 5, 13)
-
- 6. The method of claim l wherein using the text features include identifying words before and after a word gap defined by the period of silence.
-
8. A method of recognizing punctuation in computer-implemented speech recognition in a computer having a processor responsive to instructions for performing the method of recognizing punctuation, the method comprising:
-
receiving the instructions by the processor, and executing the instructions for; performing speech recognition on an utterance to produce a recognition result for the utterance; identifying a non-verbalized punctuation mark in a recognition result including predicting the non-verbalized punctuation mark using at least one acoustic feature related to the utterance; formatting the recognition result based on the identification; selecting a portion of the recognition result to be corrected that includes the non-verbalized punctuation mark; and correcting the portion of the recognition result that includes the non-verbalized punctuation mark with one of a number of correction choices, at least one of the correction choices including a change to the non-verbalized punctuation mark, the acoustic feature including an average pitch of words near the period of silence and a function of a pitch of words adjacent to the word gap, the acoustic feature based on words adjacent to the word gap including the average pitch of the words two back from the word gap and the a ratio of the average pitch of words one forward and one back from the word gap. - View Dependent Claims (9)
-
-
10. An apparatus comprising a computer-readable storage medium having instructions stored thereon that when executed by a machine result in at least the following:
-
performing speech recognition on an utterance to produce a recognition result for the utterance; identifying a non-verbalized punctuation mark in a recognition result including predicting the non-verbalized punctuation mark using at least one text feature and at least one acoustic feature related to the utterance; inserting the non-verbalized punctuation mark into the recognition result; and formatting the recognition result based on the identification of the non-verbalized punctuation mark after the non-verbalized punctuation mark has been inserted into the recognition result; wherein the acoustic feature includes one or more of a length of a period of silence and a function of pitch of words near the period of silence, the acoustic feature including an average pitch of words near the period of silence and a function of a pitch of words adjacent to the word gap, the acoustic features based on words adjacent to the word gap including the average pitch of the words two back from the word gap and the a ratio of the average pitch of words one forward and one back from the word gap.
-
-
11. A method of recognizing punctuation in computer-implemented speech recognition dictation in a computer having a processor responsive to instructions for performing the method of recognizing punctuation, the method comprising:
-
receiving the instructions by the processor, and executing the instructions for; performing speech recognition on an utterance to produce a recognition result for the utterance; identifying a non-verbalized punctuation mark in a recognition result; determining where to insert the non-verbalized punctuation mark within the recognition result based on the identification using at least one text feature and at least one acoustic feature related to the utterance to predict where to insert the non-verbalized punctuation mark; and inserting the non-verbalized punctuation mark into the recognition result; wherein the acoustic feature includes one or more of a length of a period of silence and a function of pitch of words near the period of silence, the acoustic feature including an average pitch of words near the period of silence and a function of a pitch of words adjacent to the word gap, the acoustic feature based on words adjacent to the word gap including the average pitch of the words two back from the word gap and the a ratio of the average pitch of words one forward and one back from the word gap. - View Dependent Claims (12)
-
-
14. An apparatus comprising a computer-readable storage medium having instructions stored thereon and having a processor responsive to the instructions that when executed by a machine result in at least the following:
-
receiving the instructions by the processor, and executing the instructions for; performing speech recognition on an utterance to produce a recognition result for the utterance; identifying a non-verbalized punctuation mark in a recognition result; determining where to insert the non-verbalized punctuation mark within the recognition result based on the identification using at least one text feature and at least one acoustic feature related to the utterance to predict where to insert the non-verbalized punctuation mark; and inserting the non-verbalized punctuation mark into the recognition result; wherein the acoustic feature includes one or more of a length of a period of silence and a function of pitch of words near the period of silence, the acoustic feature including an average pitch of words near the period of silence and a function of a pitch of words adjacent to the word gap, the acoustic features based on words adjacent to the word gap including the average pitch of the words two back from the word gap and the a ratio of the average pitch of words one forward and one back from the word gap.
-
Specification