Speaker-independent Voice Recognition Research Articles

A Pilot Study of an Electronic, Adolescent Version of the Quick Inventory of Depressive Symptomatology

Adolescent depression assessments are time-intensive, often requiring separate interviews with an adolescent and a parent/ informant. In adults, a self-rated, interactive voice response (IVR) version of the Quick Inventory of Depressive Symptomatology (QIDS-IVR) has been shown to be reliable, valid, and sensitive to change. An adolescent version of the QIDS (QIDS-A-IVR) was created using speaker-independent voice recognition technology. An informant version, QIDS-P-IVR, collects ratings from parents or other knowledgeable adults. The study included 27 adolescents ranging from 12 to 17 years of age, 48% of whom were female. During a single office visit, adolescents completed the QIDS-A-IVR and parents completed the QIDS-P-IVR. A clinician completed the clinician-rated adult version of the QIDS separately for adolescents (QIDS-C-A) and parents (QIDS-C-P) and the Children's Depression Rating Scale-Revised (CDRS-R). The study was conducted from October 2005 to April 2006. Cronbach alpha of the QIDS-A-IVR was .85. The QIDS-A-IVR correlated significantly with the QIDS-C-A (r = 0.95) and the CDRS-R (r = 0.76), both p < .01. Conversely, the correlations of the QIDS-A-IVR with the QIDS-P-IVR and the QIDS-C-P were small and nonsignificant. The QIDS-A-IVR required adolescents a mean of 6 minutes and 31 seconds to complete (SD = 41 seconds). The voice recognition technology correctly identified the adolescents' spoken words in 92% of the 483 spoken responses. The system recognized a response from all adolescents on all items. This study supports the reliability and validity of the QIDS-A-IVR as an adolescent depression measure. The QIDS-A-IVR may provide clinicians and researchers with a sound, technology-based method of assessing adolescent depression. Future research is needed on the informational value of parent ratings of adolescent depression.

Practical speaker‐independent voice recognition using segmental features

AbstractThis paper reports a practical method that achieves speaker‐independent large‐vocabulary voice recognition with high accuracy and high noise immunity but with small computational complexity. The first feature of the method is that highly accurate recognition is realized by using an acoustic model in which the input consists of the segmental features formed by the analysis parameters of multiple frames. The second feature is that the likelihood corresponding to the output probability of each state of the acoustic model is calculated by a linear expression with the input parameter vector as the variable. The linear expression is derived from the equal‐covariance assumption. This linear expression reduces the computational complexity and the required memory capacity even if segmental features are used, without degrading recognition performance. The third feature is that a stable word spotting function detects correct solution candidates from the noise interval by including the idea of the a posteriori probability in the likelihood calculation. This word spotting function allows voice recognition which is robust to noise to be realized. The effectiveness of the proposed system was demonstrated in a recognition experiment with superimposed interior noise of a running vehicle. © 2004 Wiley Periodicals, Inc. Electron Comm Jpn Pt 2, 87(2): 73–81, 2004; Published online in Wiley InterScience (www.interscience.wiley.com). DOI 10.1002/ecjb.10217

Simultaneous voice recognition and verification to allow access to telephone network services

The present invention describes a system and method for enabling a caller to obtain access to services via a telephone network by entering a spoken password having a plurality of digits. Preferably, the method includes the steps of: (1) prompting the caller to speak the password beginning with a first digit and ending with a last digit thereof, (2) recognizing each spoken digit of the password using a speaker-independent voice recognition algorithm, (3) following entry of the last digit of the password, determining whether the password is valid, and (4) if the password is valid, verifying the caller's identity using a voice verification algorithm.

Simultaneous speaker-independent voice recognition and verification over a telephone network

First Page

Method for recognizing alphanumeric strings spoken over a telephone network

The present invention describes a method for recognizing alphanumeric strings spoken over a telephone network (10) wherein individual character recognition need not to be uniformly high in order to achieve high string recognition accuracy. Preferably, the method uses a processing system (14) having a digital processor, an interface (42) to the telephone network (10), and a database (32) for storing a predetermined set of reference alphanumeric strings. In operation, the system (10) prompts the caller to speak the characters of a string, and characters are recognized using a speaker-independent voice recognition algorithm (48). The method calculates recognition distances between each spoken input character and the corresponding letter or digit in the same position within each reference alphanumeric string. After each character is spoken (206), captured and analyzed (208), each reference string distance is incremented (204) and the process is continued, accumulating distances for each reference string, until the last character is spoken. The reference string with the lowest cumulative distance is then declared to be the recognized string (210).

Toward Speaker Independent Voice Recognition

The purpose of this study was to investigate the extent to which off-the-shelf voice recognition equipment can perform as a speaker-independent system. By “speaker independence” is meant the ability of the system to be used by a larger number of individuals than that which trained the system (but including those who trained it), and by individuals different from those who trained it. Several independent groups of five subjects trained a threshold T600 voice recognition unit, each subject training the same 50 utterances for a total of 250 utterances. Later, a number of testing trials were conducted wherein subjects tested the system using 1) the 50 utterances that they trained, 2) the utterances that they trained plus the utterances trained by the other four subjects, and 3) the utterances trained only by the other four subjects. Measures of recognition accuracy were percent nonrecognitions and percent misrecognitions. Results will be discussed in light of additional applications for voice recognition if currently available systems prove to be able to function well in a speaker independent mode.