Speech recognition for localized content

Patent metadata

Field	Value
Publication number	US-10448115-B1
Application number	US-201615278374-A
Country	US
Kind code	B1
Filing date	Sep 28, 2016
Priority date	Sep 28, 2016
Publication date	Oct 15, 2019
Grant date	Oct 15, 2019

Abstract

A system capable of generating and storing program data associated with localized content, such as over-the-air television broadcasts. The system may update a knowledge base with the program data to improve natural language understanding (NLU) processing associated with the localized content. For example, a voice enabled device may receive a voice command and the system may perform NLU processing to interpret the voice command, determine content based on the program data and send an instruction to an output device to output the content. The program data may include information about localized content such as channels detected by the output device and/or content broadcast on the channels, including information about an output device, a channel number, call sign, station name, signal strength, etc. associated with a detected channel, and/or a program name, genre, actors, directors, producers, etc. associated with content broadcast on the channel or stored in a local storage device.

First claim

What is claimed is:

1. A computer-implemented method, the method comprising:
receiving channel data that represents a plurality of terrestrial over-the-air broadcast television channels detectable by a first device associated with a user profile, the plurality of terrestrial over-the-air broadcast television channels including a first broadcast channel;
determining, using the channel data and a program guide data source, program data including:
a list of program titles that are broadcast on the plurality of terrestrial over-the-air broadcast television channels, and
a list of broadcast time periods corresponding to each of the program titles;

receiving first audio input data from a second device associated with the user profile;
performing speech recognition on the first audio input data to determine text data;
determining, using the user profile, that the second device is associated with the first device;
configuring, using information included in the program data, a knowledge database associated with the user profile, wherein the knowledge database is used by a natural language understanding (NLU) component to interpret the text data;
determining, using the text data, the NLU component, and the knowledge database, that the text data corresponds to a command to operate the first device to output a program received using the first broadcast channel; and
sending, to the first device, an instruction to tune to the first broadcast channel and output the program.

2. The computer-implemented method of claim 1 , further comprising:
determining a channel identification associated with the first broadcast channel;
determining a call sign associated with the first broadcast channel;
determining a first program title of the program broadcast on the first broadcast channel;
determining a list of actors associated with the program;
determining a genre associated with the program; and
configuring the knowledge database to include the channel identification, the call sign, the first program title, each actor included in the list of actors, and the genre.

3. The computer-implemented method of claim 1 , wherein:
the configuring the knowledge database further comprises assigning ranking values to words included in the program data;
the determining that the text corresponds to the command further comprises:
identifying, using the NLU component a text string included in the text data,
determining, using the NLU component, that the text string is an entity,
determining, using the NLU component, that the text string is included in the knowledge database,
determining, using the NLU component and the knowledge database, a first ranking value corresponding to the text string,
determining that the first ranking value exceeds a threshold, and
determining that the text string corresponds to a title of the program.

4. The computer-implemented method of claim 1 , wherein the determining that the text data corresponds to the command further comprises:
determining that the text data includes a first word;
identifying the first word within the program data;
determining that the first word corresponds to a channel identification associated with the first broadcast channel;
determining a confidence score that the text data is associated with the program, wherein the confidence score is determined based at least in part on the first word corresponding to the channel identification; and
determining that the confidence score is above a threshold.

5. A computer-implemented method, comprising:
receiving first data indicating a plurality of content channels accessible via a first device associated with a user profile, the plurality of content channels accessible by detecting terrestrial over-the-air broadcasts and including a content channel;
determining, based on the first data, second data including information associated with the plurality of content channels;
associating the second data with the user profile;
operating a speech processing component based at least in part on the second data, wherein the speech processing component is associated with the user profile;
receiving audio data representing an utterance;
performing speech recognition on the audio data to determine text data;
determining that the audio data is associated with the first device;
performing natural language understanding, using the text data and the speech processing component, to determine that the text data corresponds to a command to operate the first device to output a first content item being broadcast on the content channel; and
sending a message to the first device to tune to the content channel.

6. The computer-implemented method of claim 5 , further comprising, based on the second data:
determining a channel identification associated with the content channel;
determining a call sign associated with the content channel;
determining a title of the first content item;
determining a list of actors associated with the first content item; and
determining a genre associated with the first content item; and
updating the user profile to include the channel identification, the call sign, the title, each actor included in the list of actors, and the genre.

7. The computer-implemented method of claim 5 , further comprising:
configuring an entity resolution component to recognize the information included in the second data; and
performing the natural language understanding, using the text data, the speech processing component, and the entity resolution component, to determine that the text data corresponds to the command to operate the first device to output the first content item.

8. The computer-implemented method of claim 5 , further comprising:
updating the user profile to assign ranking values to words included in the second data;
identifying a text string included in the text data;
determining that the text string is an entity;
determining that the text string is included in the user profile;
determining a first ranking value corresponding to the text string;
determining that the first ranking value exceeds a threshold; and
determining that the text string corresponds to a title of the first content item.

9. The computer-implemented method of claim 5 , further comprising:
receiving the audio data from the first device during a first time period;
receiving the first data from the first device during the first time period; and
sending, to the first device, the message to tune to the content channel.

10. The computer-implemented method of claim 5 , further comprising:
receiving the first data from the first device;
receiving the audio data from a second device;
determining that the second device is associated with the user profile;
determining, based at least in part on the user profile, that the audio data is associated with the first device; and
sending, to the first device, the message to tune to the content channel.

11. The computer-implemented method of claim 5 , wherein the performing natural language understanding further comprises:
determining, using the text data and the speech processing component, a first confidence score that the text data corresponds to the command;
determining, using the text data and the speech processing component, a second confidence score that the text data corresponds to a second command to operate the first device to output a second content item;
determining that the first confidence score is above a threshold;
determining that the second confidence score is above the threshold;
sending, to the first device, a second message to display a user interface listing the first content item and the second content item; and

Assignees

• Amazon Tech Inc

Inventors

• Jamal Shah Pavel
• Lawrence Mark
• Halbach Kyle Daniel
• Peacock Neal
• Quach Anderson Kailodge

Classifications

• G10L2015/088

  Word spotting · CPC title

• G10L15/16

  using artificial neural networks · CPC title

• G10L2015/223

  Execution procedure of a spoken command · CPC title

• G10L25/78

  Detection of presence or absence of voice signals (switching of direction of transmission by voice frequency in two-way loud-speaking telephone systems H04M9/10) · CPC title

• G10L15/142

  Hidden Markov Models [HMMs] · CPC title