Speech recognition method and apparatus using device information

• US 10,643,620 B2
• Filed: 05/06/2015
• Issued: 05/05/2020
• Est. Priority Date: 05/23/2014
• Status: Active Grant

• Current Assignee: Samsung Electronics Co. Ltd.
• Original Assignee: Samsung Electronics Co. Ltd.
• Inventors: Kim, Tae-yoon, Han, Chang-woo, Lee, Jae-won
• Primary Examiner(s): Kim, Jonathan C
• Application Number:
  US15/313,660
• Publication Number:
  US 20170206903A1
• Time in Patent Office: 1,826 Days
• Field of Search: None
• CPC Class Codes: G10L 15/02 Feature extraction for speech recognition; Selection of recognition unit
• CPC Class Codes: G10L 15/183 using context dependencies, e.g. language models
• CPC Class Codes: G10L 15/30 Distributed recognition, e.g. in client-server systems, for mobile phones or network applications

Abstract

A speech recognition method includes: storing at least one acoustic model (AM); obtaining, from a device located outside the ASR server, a device ID for identifying the device; obtaining speech data from the device; selecting an AM based on the device ID; performing speech recognition on the speech data by using the selected AM; and outputting a result of the speech recognition.

Claims

1. A server comprising:

a memory configured to store an acoustic model (AM) database for storing at least one AM, the AM database comprising a general AM and at least one device-adapted AM;

a receiver configured to obtain, from a device located outside the server, a device ID, which is encrypted, for identifying the device and speech data; and

a processor configured to, based on identifying that a device-adapted AM corresponding to the device ID is stored in the AM database, perform speech recognition on the speech data by using the device-adapted AM, and based on identifying that the device-adapted AM corresponding to the device ID is not stored in the AM database, perform speech recognition on the speech data by using the general AM, and output a result of the speech recognition,

wherein the memory is further configured to store a usage log database for storing a usage data log that includes the result of the speech recognition and the speech data corresponding to the result of the speech recognition,

wherein the processor is further configured to select a device ID that needs device adaptation, by monitoring the usage log database, remove speech data that is unnecessary for the device adaptation from the usage data log corresponding to the selected device ID, and generate a device-adapted AM corresponding to the selected device ID, by using speech data of the usage data log from which the speech data that is unnecessary is removed,

wherein the processor is further configured to identify that the speech data is unnecessary based on a modification record of the result of the speech recognition, and

wherein the modification record of the result of the speech recognition includes a more reliable result obtained by user'"'"'s modification upon identifying that a more reliable result of the speech recognition exists.

2. A speech recognition method comprising:

storing an acoustic model (AM) database comprising a general AM and at least one device-adapted AM;

obtaining, from a device located outside a server, a device ID, which is encrypted, for identifying the device;

obtaining speech data from the device;

based on identifying that a device-adapted AM corresponding to the device ID is stored in the AM database, performing speech recognition on the speech data by using the device-adapted AM, and based on identifying that the device-adapted AM corresponding to the device ID is not stored in the AM database, performing speech recognition on the speech data by using the general AM;

outputting a result of the speech recognition;

storing a usage data log that includes the result of the speech recognition and the speech data corresponding to the result of the speech recognition in a usage log database;

selecting a device ID that needs device adaptation, by monitoring the usage log database and removing speech data that is unnecessary for the device adaptation from the usage data log corresponding to the selected device ID; and

generating a device-adapted AM corresponding to the selected device ID, by using speech data of the usage data log from which the speech data that is unnecessary is removed,

wherein the removing the speech data that is unnecessary comprises identifying that the speech data is unnecessary based on a modification record of the result of the speech recognition, and

wherein the modification record of the result of the speech recognition includes a more reliable result obtained by user'"'"'s modification upon identifying that a more reliable result of the speech recognition exists.

3. A non-transitory computer-readable recording storage medium having stored thereon a computer program, which when executed by a computer, performs the method of claim 2.

4. A device comprising:

a memory configured to store a device ID, which is encrypted, for identifying a device; and

at least one processor configured to execute instructions stored in the memory to implement;

an input interface configured to obtain an input of a speech for speech recognition;

a speech generator configured to generate speech data by processing the speech;

a transmitter configured to transmit the device ID and the speech data to an automatic speech recognition (ASR) server; and

a receiver configured to obtain a result of the speech recognition performed on the speech data from the ASR server, the result of the speech recognition comprising an identification of whether a device-adapted acoustic model (AM) corresponding to the device ID is stored in an acoustic model AM database in the server,

wherein the speech recognition further comprises selecting a device ID that needs device adaptation, by monitoring a usage log database, remove speech data unnecessary for the device adaptation from the usage data log corresponding to the selected device ID, and generating a device-adapted AM corresponding to the selected device ID, by using speech data of the usage data log from which the unnecessary speech data is removed,

wherein the processor is further configured to identify that the speech data is unnecessary based on a modification record of the result of the speech recognition, and

wherein the modification record of the result of the speech recognition includes a more reliable result obtained by user'"'"'s modification upon identifying that a more reliable result of the speech recognition exists.

5. The device of claim 4, wherein the at least one processor is further configured to extract data used for the speech recognition, as speech data.

6. The device of claim 4, wherein the device has a plurality of device identifications (IDs).

7. The device of claim 6, wherein a setting of the device varies depending on the device IDs.

8. The device of claim 4, wherein the transmitter is further configured to transmit location information of the device.