Speaker verification based on acoustic behavioral characteristics of the speaker

• US 9,659,564 B2
• Filed: 10/22/2015
• Issued: 05/23/2017
• Est. Priority Date: 10/24/2014
• Status: Active Grant

• Current Assignee: Sestek Ses Ve Iletisim Bilgisayar Teknolojileri
• Original Assignee: Sestek Ses Ve Iletisim Bilgisayar Teknolojileri
• Inventors: Arslan, Mustafa Levent
• Primary Examiner(s): Sirjani, Fariba
• Application Number:
  US14/919,750
• Publication Number:
  US 20160118050A1
• Time in Patent Office: 579 Days
• CPC Class Codes: G10L 15/07 to the speaker
• CPC Class Codes: G10L 15/26 Speech to text systems G10L15/08 takes precedence
• CPC Class Codes: G10L 17/26 Recognition of special voice characteristics, e.g. for use in lie detectors; Recognition of animal voices
• CPC Class Codes: G10L 2015/227 of the speaker; Human-factor methodology
• CPC Class Codes: G10L 25/63 for estimating an emotional state

Abstract

The present invention relates to a non-standard speech detection system and method whereby a speech is analyzed based on models that are trained using personalized speech for each individual. The model is stored in a database and used to analyze a speech in real time to determine the content and behavior of an individual who is a party to a conversation that produces the speech. The results of the analysis can be used to determine if a conversation takes place under normal circumstances or under extraneous circumstances.

Claims

1. A non-standard speech detection system that is capable of analyzing human voice to detect the speech style of an individual for verifying identity of the individual according to his speech style, the system comprising:

at least one voice recording unit for recording personal standard speeches of the individual in his/her normal life;

at least one voice database wherein the personal standard speeches recorded by the voice recording unit are stored as voice recordings;

at least one acoustic examination module for determining acoustic speech characteristics of the individual from the voice recordings stored in the voice database;

at least one speech recognition module for determining behavioral characteristics and speech content characteristics of the individual from the voice recordings stored in the voice database and capable of making speech analysis by converting the voice recordings into text;

at least one analysis module for creating a personal standard speech style model by examining the acoustic speech characteristics and the behavioral speech characteristics of the individual determined by the acoustic examination module and the speech content characteristics of the individual determined by the speech recognition module;

at least one personal standard speech style model database wherein the personal standard speech style models created by the analysis module are recorded in a personalized way;

at least one evaluation module for comparing a new voice input with the recorded personal standard speech style models;

at least one transmission unit capable of transmitting result reached after the comparison of the evaluation module to related units,

wherein the speech content characteristics of the individual include sentence patterns, words and phrases, language model n-gram parameters, and semantic parameters obtained from the text,

wherein the evaluation module evaluates whether speech content characteristics, acoustic speech characteristics, and behavioral speech characteristics of the new voice input match the personal standard speech style model of the individual and verifies identity of a caller producing the new voice if a match occurs,

wherein the behavioral speech characteristics of the individual include speech behavioral characteristic parameters of the individual, interruption interval for another party to the conversation, and speech overlap with the other party, and

wherein the speech behavioral characteristic parameters include speaking speed, monotony, hesitance, and loudness of voice.

2. The non-standard speech detection system according to claim 1, wherein the voice recording unit records standard speeches of the individual by means of recording conversations made over voice communication channels and/or requesting voice recording from the individual.

3. The non-standard speech detection system according to claim 1, wherein the personal standard speeches collected by the voice recording unit are recorded in a personalize way individually.

4. The non-Standard speech detection system according to claim 1, wherein the related pre-determined places include financial and e-commerce institutions that require verification of the caller.

5. The non-standard speech detection system according to claim 1, wherein the evaluation module determines whether the speech is performed in a stand/non-standard way by comparing the new voice inputs taken in the individual'"'"'s communications made over a voice communication channel with the personal speech style models of the individual recorded in the standard speech style model database.

6. The non-standard speech detection system according to claim 1,

wherein the transmission unit provides feedback by sending a message with e-mail, SMS, or pop-up methods and reflecting information on a screen of an electronic device of a related entity requesting the verification of the caller over an interface, and

wherein the electronic device is selected from a group consisting of PC, tablet and smart phone.

7. A non-standard speech detection method for verifying identity of a caller according to his speech, the method comprising the steps of:

recording, to a database, conversations of an individual made by the individual over a voice communication channel to obtain voice recordings including voice of the individual;

analyzing acoustic speech characteristics from the voice recordings in the voice database to obtain speech characteristics of the individual;

analyzing behavioral characteristics of the individual by using a speech recognition module applied to the voice recordings in the voice database to obtain behavioral characteristics of the individual;

converting the voice recordings in the voice database into texts and analyzing the speech content of the individual from the texts;

creating a personal standard speech style from the speech characteristics, the behavioral characteristics, and the speech content;

storing the personal standard speech style in a standard speech style model database;

comparing acoustic speech characteristics and content of a new voice input from a caller with the personal standard speech style in the standard speech style model database;

evaluating whether the new voice matches the personal standard speech style;

verifying identity of the caller according to results of the evaluating to obtain a verification of the identity of the caller;

transmitting the the verification of the identity of the user to related pre-determined places,

wherein the step of analyzing speech content of the individual includes obtaining sentence patterns, words and phrases, language model n-gram parameters, and semantic parameters from the content of the text,

wherein the step of analyzing behavioral characteristics of the individual by using the speech recognition module applied to the voice recordings in the voice database comprises;

evaluating speech behavioral characteristic parameters of the individual, interruption interval for another party to the conversation, and speech overlap with the other party,

wherein the speech behavioral characteristic parameters include speaking speed, monotony, hesitance, and loudness of voice.

8. The non-standard speech detection method according to claim 7, the step of analyzing the acoustic speech characteristics from the voice recordings in the voice database, comprises:

using acoustic parameters, which are selected from a group consisting of prosodic parameters, pitch frequency, emphasis parameters, energy, spectrum parameters, duration parameters, Mel Frequency Cepstral Coefficients, harmonics, jitter and shimmer.

9. The non-standard speech detection method according to claim 7, wherein at the step of “

creating a personal standard speech style from the analysis results of the individual'"'"'s speech characteristics, speech content and behavioral characteristics”

;

acoustic speech characteristics, speech content and behavioral characteristics are obtained from system modules.

10. The non-standard speech detection method according to claim 7, wherein the related pre-determined places include financial and e-commerce institutions that require verification of the caller.

11. The non-standard speech detection method according to claim 7, wherein a transmission unit provides feedback by sending message with e-mail, SMS, pop-up methods and reflecting information on screen of an electronic device of the individual over an interface at the step of “

transmitting the information obtained after the evaluation process carried out by the evaluation module to related pre-determined places”

, the electronic device is selected from a group consisting of PC, tablet and smart phone.