Hello everyone,
In this project, I must create a Matlab program that can recognize several names that have been spoken and recorded.
In my main program, I have "cropped" the voice samples of "dead space" so that the data behind them is easier to filter.
Then, I have computed two features (numbers) from each voice sample. These features will be used to distinguish among the voices. The features are designed from a spectogram of the voices. Since the overall energy of each signal may change depending on how loud the name is said, I have used ratios to make my features independent of overall energy.
So feature 1 is the ratio of the average energy in the high frequencies in the first half of the spectogram to the low frequencies in the first half.
Feature 2 is the ratio of the average energy in th high frequencies in the last half to the the low frequencies in the last half.
Now that I have computed the features for each voice sample, I need to devise a decision rule that will associate any possible pair of features given to one of the voice samples.
Can anyone offer suggestions on how to make a decision rule and/or how to apply the rule to the voice samples so that it either recognizes the voice or doesnt?
Thanks
-Azi
In this project, I must create a Matlab program that can recognize several names that have been spoken and recorded.
In my main program, I have "cropped" the voice samples of "dead space" so that the data behind them is easier to filter.
Then, I have computed two features (numbers) from each voice sample. These features will be used to distinguish among the voices. The features are designed from a spectogram of the voices. Since the overall energy of each signal may change depending on how loud the name is said, I have used ratios to make my features independent of overall energy.
So feature 1 is the ratio of the average energy in the high frequencies in the first half of the spectogram to the low frequencies in the first half.
Feature 2 is the ratio of the average energy in th high frequencies in the last half to the the low frequencies in the last half.
Now that I have computed the features for each voice sample, I need to devise a decision rule that will associate any possible pair of features given to one of the voice samples.
Can anyone offer suggestions on how to make a decision rule and/or how to apply the rule to the voice samples so that it either recognizes the voice or doesnt?
Thanks
-Azi