# US 10,504,503 B2 · Method and apparatus for recognizing speech > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 10,504,503 B2 - Document ID: US10504503B2 - Status: Granted - Kind: B2 - Application number: 15/585,334 - Filed: 2017-05-03 - Published: 2018-06-14 - Granted: 2019-12-10 - Earliest claimed: 2016-12-14 - Assignees: Samsung Electronics Co., Ltd. - Applicants: Samsung Electronics Co., Ltd. - Inventors: Hodong Lee; Hoshik Lee - Web page: https://patentgrep.com/patent/US10504503B2 ## CPC - G10L 15/02: Feature extraction for speech recognition; Selection of recognition unit - G10L 15/20: Speech recognition techniques specially adapted for robustness in adverse environments, e.g. in noise, of stress induced speech (G10L21/02 takes precedence) - G10L: SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING ## Abstract A method performed by a speech recognizing apparatus to recognize speech includes: obtaining a distance from the speech recognizing apparatus to a user generating a speech signal; determining a normalization value for the speech signal based on the distance; normalizing a feature vector extracted from the speech signal based on the normalization value; and performing speech recognition based on the normalized feature vector. ## Claims ### Claim 1 A processor-implemented method performed by a speech recognizing apparatus to recognize speech, the method comprising: obtaining a distance from the speech recognizing apparatus to a user generating a speech signal; determining a reference distance being closest to the obtained distance from among a plurality of reference distances; determining a normalization value for the speech signal based on one or more reference normalization values obtained from a database using the determined reference distance, the database storing a plurality of reference normalization values specific to each of the plurality of reference distances, wherein the determining of the normalization value comprises estimating a normalization value corresponding to the obtained distance based on the one or more reference normalization values of the plurality of reference normalization values, and interpolating the normalization value corresponding to the obtained distance based on a result of approximating the reference normalization values; performing a respective extracting of feature vectors from the speech signal; performing a normalizing of the extracted feature vectors based on the determined normalization value; and performing speech recognition based on the normalized feature vectors. ### Claim 2 (depends on 1) The method of claim 1, wherein the determining of the normalization value comprises determining from the database, as the normalization value, a reference normalization value, from the plurality of reference normalization values, corresponding to the determined reference distance. ### Claim 3 (depends on 1) The method of claim 1, wherein the normalization value comprises an average value and a variance value for converting an average of n-th components included in the feature vectors of the speech signal into 0 and converting a variance of the n-th components into a unit variance value. ### Claim 4 (depends on 1) The method of claim 1, wherein the performing of the normalizing of the feature vectors comprises performing of the normalizing of the feature vectors based on a cepstral mean variance normalization (CMVN) scheme based on the normalization value. ### Claim 5 (depends on 1) The method of claim 1, wherein the determining of the normalization value comprises determining from the database, as the normalization value, a reference normalization value, from the plurality of reference normalization values, corresponding to the obtained distance and a position of the speech recognizing apparatus. ### Claim 6 (depends on 1) The method of claim 1, further comprising: updating reference normalization values, of the plurality of reference normalization values, corresponding to the determined reference distance based on the normalized feature vectors. ### Claim 7 (depends on 1) The method of claim 1, wherein the obtaining of the distance comprises obtaining the distance based on an output value of an image sensor configured to photograph the user or an output value of an infrared sensor configured to detect a time of flight of an infrared ray projected towards the user and reflected by the user. ### Claim 8 (depends on 1) The method of claim 1, wherein the obtaining of the distance comprises estimating the distance based on the speech signal. ### Claim 9 (depends on 1) A non-transitory computer-readable storage medium storing instructions that, when executed by a processor, cause the processor to perform the method of claim 1. ### Claim 10 A speech recognizing apparatus comprising: a processor configured to: obtain a distance from the speech recognizing apparatus to a user generating a speech signal; determine a reference distance being closest to the obtained distance from among a plurality of reference distances; determine a normalization value for the speech signal based on one or more reference normalization values obtained from a database using the determined reference distance, the database storing a plurality of reference normalization values specific to each of the plurality of reference distances; perform a respective extracting of feature vectors from the speech signal; perform a normalizing of the extracted feature vectors based on the determined normalization value; and perform speech recognition based on the normalized feature vectors, wherein, for the determining of the normalization value, the processor is further configured to estimate a normalization value corresponding to the obtained distance based on the one or more reference normalization values of the plurality of reference normalization values, and interpolate the normalization value corresponding to the obtained distance based on a result of approximating the reference normalization values. ### Claim 11 (depends on 10) The speech recognizing apparatus of claim 10, wherein the processor is further configured to determine, as the normalization value, a reference normalization value, from the plurality of reference normalization values, corresponding to the determined reference distance. ### Claim 12 (depends on 10) The speech recognizing apparatus of claim 10, wherein the normalization value comprises an average value and a variance value for converting an average of n-th components included in the feature vectors of the speech signal into 0 and converting a variance of the n-th components into a unit variance value. ### Claim 13 (depends on 10) The speech recognizing apparatus of claim 10, wherein the processor is further configured to perform the normalizing of the feature vectors based on a cepstral mean variance normalization (CMVN) scheme based on the normalization value. ### Claim 14 (depends on 10) The speech recognizing apparatus of claim 10, wherein the processor is further configured to obtain the distance based on an output value of an image sensor configured to photograph the user or an output value of an infrared sensor configured to detect a time of flight of an infrared ray projected towards the user and reflected by the user. ### Claim 15 (depends on 10) The speech recognizing apparatus of claim 10, wherein the processor is further configured to estimate the distance based on the speech signal. ### Claim 16 (depends on 10) The speech recognizing apparatus of claim 10, further comprising: the memory, the memory further comprising at least one instruction executable by the processor, wherein the processor is configured to, in response to the at least one instruction being executed by the processor, perform the obtaining of the distance, the determining of the reference distance, the determining of the normalization value, the performing of the respective extracting, the performing of the normalizing, and the performing of the speech recognition. ### Claim 17 (depends on 10) The speech recognizing apparatus of claim 10, wherein the speech recognizing apparatus is an electronic device comprising an image sensor configured to photograph the user. ### Claim 18 A speech recognizing apparatus, comprising: a processor configured to obtain a distance from the speech recognizing apparatus to a user generating a speech signal; perform a respective extracting of feature vectors from the speech signal; determine a reference distance being closest to the obtained distance from among a plurality of reference distances; perform a normalizing of the extracted feature vectors based on a normalization value based on one or more reference normalization values obtained from a memory using the determined reference distance, the memory storing a plurality of normalization values specific to each of the plurality of reference distances, wherein, for the performing the normalizing, the processor is further configured to estimate a normalization value corresponding to the obtained distance based on the one or more reference normalization values of the plurality of reference normalization values, and interpolate the normalization value corresponding to the obtained distance based on a result of approximating the reference normalization values; and perform speech recognition based on the normalized feature vectors. ### Claim 19 (depends on 18) The speech recognizing apparatus of claim 18, further comprising: the memory, wherein the processor is configured to perform the normalizing of the feature vectors by applying a reference normalization value, among the plurality of reference normalization values, corresponding to the determined reference distance. ### Claim 20 (depends on 19) The speech recognizing apparatus of claim 19, wherein the processor is further configured to update reference normalization values, of the plurality of reference normalization values, based on the normalized feature vectors. ### Claim 21 (depends on 18) The speech recognizing apparatus of claim 18, wherein the processor is configured to perform the normalizing of the feature vectors based on a cepstral mean variance normalization scheme.