×

Training punctuation models

  • US 9,135,231 B1
  • Filed: 01/31/2013
  • Issued: 09/15/2015
  • Est. Priority Date: 10/04/2012
  • Status: Active Grant
First Claim
Patent Images

1. A computer-implemented method comprising:

  • receiving audio data encoding an utterance of a user;

    obtaining, by an automated speech recognizer, a transcription of the utterance of the user;

    determining, by one or more processors, that one or more punctuation rules dictate that a punctuation mark may optionally be placed at a particular location within the transcription;

    identifying, by the one or more processors, text that (i) was written by the user, (ii) is similar to the transcription, and (iii) includes a location that corresponds to the particular location within the transcription;

    determining, by the one or more processors, whether the user placed the punctuation mark at the corresponding location within the text that was written by the user, and that is similar to the transcription; and

    determining, by the one or more processors, whether to place or not place the punctuation mark at the particular location within the transcription of the utterance, according to the one or more punctuation rules, based on determining whether the user placed or did not place the punctuation mark at the corresponding location within the text that was written by the user, and that is similar to the transcription;

    creating an electronic communication using the punctuated transcription, wherein the electronic communication is a type of electronic communication selected from the group consisting of an electronic mail message, a SMS text message, a social media message or post, or a word processing document; and

    transmitting, by a client computing device, the electronic communication to a communication recipient.

View all claims
  • 2 Assignments
Timeline View
Assignment View
    ×
    ×