Source-Filter Model
Definition
Standard model of voice production: the glottal source (vocal-fold vibration) is filtered by the vocal-tract cavity to produce the speech signal.
- Components
- Glottal source and vocal-tract filter
- Source
- Vocal-fold vibration
- Filter
- Vocal-tract cavity resonances
- Field
- Standard model of voice production in forensic phonetics
Common questions
Why is the source-filter model central to forensic speaker comparison?+
It separates speaker-specific traits into two categories, the glottal source features like fundamental frequency that reflect vocal-fold physiology, and the filter features like formant frequencies that reflect vocal-tract shape, letting examiners select and quantify features most likely to carry individualising information rather than treating the voice as one undifferentiated signal.
Can the source and filter components be manipulated independently, and what does that mean for voice disguise?+
Yes, a speaker can alter pitch, largely a source property, while vocal-tract shape changes such as mouth position alter filter properties, so deliberate voice disguise often shifts one component more than the other, and examiners look at whether disguised and known samples remain consistent in the harder-to-control component.
Related terms
- FASR
- Forensic Automatic Speaker Recognition. Likelihood-ratio framework using statistical speaker models (GMM, i-vectors, x-vectors).
- Formants (F1, F2, F3, F4)
- Resonant frequencies of the vocal tract that shape vowel quality. F1 and F2 together distinguish vowels (for example /i/ has low F1...
- Fundamental Frequency (F0)
- Rate of vocal-fold vibration during voiced speech, perceived as pitch. Typical adult male 80 to 180 Hz, adult female 165 to 255...
- Likelihood Ratio (LR)
- The ratio of two conditional probabilities: the probability of the observed evidence given the prosecution's hypothesis (same source), divided by the probability...
- MFCC
- Mel-Frequency Cepstral Coefficients. The dominant acoustic feature set for automatic speaker recognition.
- Voice Spectrogram (Sonogram)
- Time-frequency plot of speech. Wideband (about 300 Hz analysis bandwidth) shows formants; narrowband (about 45 Hz) shows harmonics.
- Voiceprint
- Lawrence Kersta's 1962 term for the claim that the spectrographic pattern of an individual's speech is unique enough to serve as a...