Context #
Learned in this study #
Things to explore #
- MFC/MFCC
- Recognize speakers using per speaker models
- Start with a single model for all speakers, and slowly figure out when a given speaker speaks, then retrain their individual model to become more and more accurate
To try #
- Determine vocal tract size
Overview #
The goal of this project is to recognize a person based on a record of his/her voice.
- Record audio
- Convert a certain window size (e.g., 20ms long) of the signal into the frequency domain using a fast Fourier transform
From the Deep Learning Book #
- Factors of variation
- Age
- Sex
- Accent
- Words spoken
From the Bay Area Deep Learning School #
- Application matters
- Styles of speech
- Read
- Conversational
- Spontaneous
- Command/control
- Issues
- Disfluency/Stuttering
- Noise
- Microphone quality/Number of channels
- Far field
- Reverb/Echo
- Lombard effect
- Speaker accents
- Styles of speech
From PHO121 - Speech Analysis #
- Vowels can be classified by their two first formants (F1 and F2)
- Resonant frequencies that can roughly be associated with the size of specific cavities in the vocal tract
- F1: Pharyngeal cavity
- F2: Front cavity