Detecting emotions using voice signal analysis

• US 7,222,075 B2
• Filed: 07/12/2002
• Issued: 05/22/2007
• Est. Priority Date: 08/31/1999
• Status: Expired due to Term

• Current Assignee: Accenture Global Services Limited (Accenture PLC)
• Original Assignee: Accenture LLP (Accenture PLC)
• Inventors: Petrushin, Valery A.
• Primary Examiner(s): Hudspeth; David
• Assistant Examiner(s): Sked; Matthew J.
• Application Number:
  US10/194,908
• Publication Number:
  US 20020194002A1
• Time in Patent Office: 1,775 Days
• Field of Search: None
• US Class Current: 704/270
• 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 25/30 using neural networks
• CPC Class Codes: H04M 3/436 Arrangements for screening incoming calls , i.e. evaluating the characteristics of a call before deciding whether to answer it based on the calling party profile H04M3/42059; based on location H04M3/42348; based on presence H04M3/42365; diversion H04M3/54
• CPC Class Codes: H04M 3/533 Voice mail systems

Abstract

A system and method are provided for detecting emotional states using statistics. First, a speech signal is received. At least one acoustic parameter is extracted from the speech signal. Then statistics or features from samples of the voice are calculated from extracted speech parameters. The features serve as inputs to a classifier, which can be a computer program, a device or both. The classifier assigns at least one emotional state from a finite number of possible emotional states to the speech signal. The classifier also estimates the confidence of its decision. Features that are calculated may include a maximum value of a fundamental frequency, a standard deviation of the fundamental frequency, a range of the fundamental frequency, a mean of the fundamental frequency, and a variety of other statistics.

Claims

1. A method of detecting an emotional state of a telephone caller or voice mail message, the method comprising:

providing a speech signal from a telephone call or voice mail message;

dividing the speech signal into at least one of segments, frames, and subframes;

extracting at least one acoustic feature from the speech signal;

calculating statistics from the at least one acoustic feature;

classifying the speech with at least one neural network classifier as belonging to at least one emotional state; and

storing in memory and outputting in a human-recognizable format an indication of the at least one emotional state,

wherein the speech is classified by a classifier taught to recognize at least one emotional state from a finite number of emotional states,

wherein the indication is a probability of at least one emotion.

2. A method of detecting an emotional state of a telephone caller or voice mail message, the method comprising:

providing a speech signal from a telephone call or voice mail message;

dividing the speech signal into at least one of segments, frames, and subframes;

extracting at least one acoustic feature from the speech signal;

calculating statistics from the at least one acoustic feature;

classifying the speech with at least one neural network classifier as belonging to at least one emotional state; and

storing in memory and outputting in a human-recognizable format an indication of the at least one emotional state,

wherein the speech is classified by a classifier taught to recognize at least one emotional state from a finite number of emotional states,

wherein the indication is a statistic of at least one feature of a voice signal.

3. The method of claim 2, wherein the statistic is selected from the group consisting of a maximum value of a fundamental frequency, a standard deviation of the fundamental frequency, a range of the fundamental frequency, a mean of the fundamental frequency, a mean of a bandwidth of a first formant, a mean of a bandwidth of a second formant, an energy standard deviation, a speaking rate, a slope of a fundamental frequency, a maximum of a first formant, an energy maximum, an energy range, a first formant range, a second formant range, a percentage of silent time of at least one segment of the voice signal, a length of the voice signal, and a distribution of emotions in the voice signal.

4. A system for classifying speech contained in a telephone call or voice mail message, the system comprising:

a computer system comprising a central processing unit, an input device, at least one memory for storing data indicative of a speech signal, and an output device;

logic for receiving and analyzing a speech signal of a telephone call or voice mail message;

logic for dividing the speech signal of the telephone call or voice mail message;

logic for extracting at least one feature from the speech signal of the telephone call or voice mail message;

logic for calculating statistics of the speech of the telephone call or voice mail message;

logic for at least one neural network for classifying the speech of the telephone call or voice mail message as belonging to at least one of a finite number of emotional states; and

logic for storing in memory and outputting an indication of the at least one emotional state of the telephone caller or voice mail message,

wherein the indication is a probability of at least one emotion.

5. A system for classifying speech contained in a telephone call or voice mail message, the system comprising:

a computer system comprising a central processing unit, an input device, at least one memory for storing data indicative of a speech signal, and an output device;

logic for receiving and analyzing a speech signal of a telephone call or voice mail message;

logic for dividing the speech signal of the telephone call or voice mail message;

logic for extracting at least one feature from the speech signal of the telephone call or voice mail message;

logic for calculating statistics of the speech of the telephone call or voice mail message;

logic for at least one neural network for classifying the speech of the telephone call or voice mail message as belonging to at least one of a finite number of emotional states; and

logic for storing in memory and outputting an indication of the at least one emotional state of the telephone caller or voice mail message,

wherein the indication is a statistic of at least one feature of a voice signal.

6. The system of claim 5, wherein the statistic is selected from the group consisting of a maximum value of a fundamental frequency, a standard deviation of the fundamental frequency, a range of the fundamental frequency, a mean of the fundamental frequency, a mean of a bandwidth of a first formant, a mean of a bandwidth of a second formant, an energy standard deviation, a speaking rate, a slope of a fundamental frequency, a maximum of a first formant, an energy maximum, an energy range, a first formant range, a second formant range, a percentage of silent time of at least one segment of the voice signal, a length of the voice signal, and a distribution of emotions in the voice signal.

7. A method of recognizing emotional states in a voice of a telephone call or voice mail message, the method comprising:

providing a first plurality of voice samples;

obtaining a second plurality of voice samples of a telephone caller, from a telephone call or voice mail message;

identifying each sample of said pluralities of samples as belonging to a predominant emotional state;

dividing each sample into at least one of frames, subframes, and segments;

extracting at least one acoustic feature for each sample of the pluralities of samples;

calculating statistics of the speech samples from the at least one feature;

classifying an emotional state in the first plurality of samples with at least one neural network;

training the at least one neural network to recognize an emotional state from the statistics by comparing the results of identifying and classifying for the first plurality of samples;

classifying an emotion in the second plurality of voice samples obtained from a telephone call or voice mail message with the at least one trained neural network; and

storing in memory and outputting in a human-recognizable format an indication of the emotional state of the telephone caller or voice mail message,

wherein the indication is a probability of at least one emotion.

8. A method of recognizing emotional states in a voice of a telephone call or voice mail message, the method comprising:

providing a first plurality of voice samples;

obtaining a second plurality of voice samples of a telephone caller, from a telephone call or voice mail message;

identifying each sample of said pluralities of samples as belonging to a predominant emotional state;

dividing each sample into at least one of frames, subframes, and segments;

extracting at least one acoustic feature for each sample of the pluralities of samples;

calculating statistics of the speech samples from the at least one feature;

classifying an emotional state in the first plurality of samples with at least one neural network;

training the at least one neural network to recognize an emotional state from the statistics by comparing the results of identifying and classifying for the first plurality of samples;

classifying an emotion in the second plurality of voice samples obtained from a telephone call or voice mail message with the at least one trained neural network; and

storing in memory and outputting in a human-recognizable format an indication of the emotional state of the telephone caller or voice mail message,

wherein the indication is a statistic of at least one feature of a voice signal.

9. The method of claim 8, wherein the statistic is selected from the group consisting of a maximum value of a fundamental frequency, a standard deviation of the fundamental frequency, a range of the fundamental frequency, a mean of the fundamental frequency, a mean of a bandwidth of a first formant, a mean of a bandwidth of a second formant, an energy standard deviation, a speaking rate, a slope of a fundamental frequency, a maximum of a first formant, an energy maximum, an energy range, a first formant range, a second formant range, a percentage of silent time of at least one segment of the voice signal, a length of the voice signal, and a distribution of emotions in the voice signal.

10. A system for detecting an emotional state of a telephone caller or voice mail message from a voice signal of a telephone call or voice mail message, the system comprising:

a speech reception device;

at least one computer connected to the speech reception device;

at least one memory operably connected to the at least one computer;

a computer program including at least one neural network for dividing the voice signal into a plurality of segments, and for analyzing the segments according to features of the segments to detect the emotional state in the voice signal;

a database of speech signal features and statistics accessible to the computer for comparison with features of the voice signal; and

an output device coupled to the computer for notifying a user of the emotional state of the telephone caller or voice mail message detected in the voice signal,

wherein the indication is a probability of at least one emotion.

11. A system for detecting an emotional state of a telephone caller or voice mail message from a voice signal of a telephone call or voice mail message, the system comprising:

a speech reception device;

at least one computer connected to the speech reception device;

at least one memory operably connected to the at least one computer;

a computer program including at least one neural network for dividing the voice signal into a plurality of segments, and for analyzing the segments according to features of the segments to detect the emotional state in the voice signal;

a database of speech signal features and statistics accessible to the computer for comparison with features of the voice signal; and

an output device coupled to the computer for notifying a user of the emotional state of the telephone caller or voice mail message detected in the voice signal,

wherein the indication is a statistic of at least one feature of a voice signal.

12. The system of claim 11, wherein the statistic is selected from the group consisting of a maximum value of a fundamental frequency, a standard deviation of the fundamental frequency, a range of the fundamental frequency, a mean of the fundamental frequency, a mean of a bandwidth of a first formant, a mean of a bandwidth of a second formant, an energy standard deviation, a speaking rate, a slope of a fundamental frequency, a maximum of a first formant, an energy maximum, an energy range, a first formant range, a second formant range, a percentage of silent time of at least one segment of the voice signal, a length of the voice signal, and a distribution of emotions in the voice signal.