Shortlist B: a Bayesian model of continuous speech recognition.
Dennis Norris1, James M McQueen
1Medical Research Council, Cognition and Brain Sciences Unit, Cambridge, UK. dennis.norris@mrc-cbu.cam.ac.uk
Psychological Review
|April 23, 2008
Summary
This study introduces Shortlist B, a Bayesian model for continuous speech recognition. It suggests listeners use optimal Bayesian decisions for spoken word recognition, accounting for segmentation and frequency effects.
Area of Science:
- Cognitive Science
- Computational Linguistics
- Psychology
Background:
- Existing models of speech recognition often use connectionist principles.
- Previous models like Shortlist relied on discrete phoneme inputs and lacked Bayesian foundations.
Purpose of the Study:
- To present a novel Bayesian model (Shortlist B) for continuous speech recognition.
- To investigate how listeners segment continuous speech and recognize words.
- To explore the role of Bayesian principles in spoken word recognition.
Main Methods:
- Developed a Bayesian model, Shortlist B, incorporating parallel lexical hypothesis evaluation.
- Utilized phonologically abstract representations and a feedforward architecture.
- Input data consisted of phoneme probabilities over time, derived from a large-scale auditory gating study.
Main Results:
- The Shortlist B model successfully accounts for continuous speech segmentation data.
- The model explains word frequency effects and the impact of mispronunciations on recognition.
- Simulations support the model's ability to explain lexical involvement in phonemic decisions.
Conclusions:
- Listeners appear to make optimal Bayesian decisions during spoken word recognition.
- The Bayesian approach provides a powerful framework for modeling speech perception.
- Shortlist B offers a significant advancement over previous interactive-activation models.
Related Concept Videos
01:30Classification of Signals
In signal processing, signals are classified based on various characteristics: continuous-time versus discrete-time, periodic versus aperiodic, analog versus digital, and causal versus noncausal. Each category highlights distinct properties crucial for understanding and manipulating signals.
A continuous-time signal holds a value at every instant in time, representing information seamlessly. In contrast, a discrete-time signal holds values only at specific moments, often denoted as x(n), where...
A continuous-time signal holds a value at every instant in time, representing information seamlessly. In contrast, a discrete-time signal holds values only at specific moments, often denoted as x(n), where...
01:24BIBO stability of continuous and discrete -time systems
System stability is a fundamental concept in signal processing, often assessed using convolution. For a system to be considered bounded-input bounded-output (BIBO) stable, any bounded input signal must produce a bounded output signal. A bounded input signal is one where the modulus does not exceed a certain constant at any point in time.
To determine the BIBO stability, the convolution integral is utilized when a bounded continuous-time input is applied to a Linear Time-Invariant (LTI) system.
To determine the BIBO stability, the convolution integral is utilized when a bounded continuous-time input is applied to a Linear Time-Invariant (LTI) system.
01:24One-Compartment Open Model: Wagner-Nelson and Loo Riegelman Method for ka Estimation
This lesson introduces two critical methods in pharmacokinetics, the Wagner-Nelson and Loo-Riegelman methods, used for estimating the absorption rate constant (ka) for drugs administered via non-intravenous routes. The Wagner-Nelson method relates ka to the plasma concentration derived from the slope of a semilog percent unabsorbed time plot. However, it is limited to drugs with one-compartment kinetics and can be impacted by factors like gastrointestinal motility or enzymatic degradation.
On...
On...
01:19Expected Frequencies in Goodness-of-Fit Tests
A goodness-of-fit test is conducted to determine whether the observed frequency values are statistically similar to the frequencies expected for the dataset. Suppose the expected frequencies for a dataset are equal such as when predicting the frequency of any number appearing when casting a die. In that case, the expected frequency is the ratio of the total number of observations (n) to the number of categories (k).
01:03Prediction Intervals
The interval estimate of any variable is known as the prediction interval. It helps decide if a point estimate is dependable.
However, the point estimate is most likely not the exact value of the population parameter, but close to it. After calculating point estimates, we construct interval estimates, called confidence intervals or prediction intervals. This prediction interval comprises a range of values unlike the point estimate and is a better predictor of the observed sample value, y.
The...
However, the point estimate is most likely not the exact value of the population parameter, but close to it. After calculating point estimates, we construct interval estimates, called confidence intervals or prediction intervals. This prediction interval comprises a range of values unlike the point estimate and is a better predictor of the observed sample value, y.
The...
01:11Sampling Continuous Time Signal
In signal processing, a continuous-time signal can be sampled using an impulse-train sampling technique, followed by the zero-order hold method. Impulse-train sampling involves the use of a periodic impulse train, which consists of a series of delta functions spaced at regular intervals determined by the sampling period. When a continuous-time signal is multiplied by this impulse train, it generates impulses with amplitudes corresponding to the signal's values at the sampling points.
In the...
In the...