Development of the 1D Digital Waveguide Vocal Tract Modelling

Code Design

Clicking on the IPA representation will link to the examples further down the page.

---

Development

When developing the model, the place to start would be looking at the excitement of the system. The system allows Liljencrants-Fant (LF) glottal flow model to synthesise the voice, the system reads in the two LF inputs, one with vibrato and one without, provided in the lab and uses it as excitement for the system. In addition to the LF inputs, the system allows 1 second of noise to be used as an excitement, below are the experiments with different types of noise, unit impulse and LFs. The phomeme being synthesised is a.

After completing the lab and implementing the different excitement types, the next stage was to focus on was creating plosives. Plosives are as as simple as stopping the sound midway through creating it. The code works by creating an envelope that goes from 0 to 1 over 1/8th of the total samples. This envelope can then be inserted at any point during the system with ones filling in the remaining space. The mouth end is then updated with the value of the envelope multiplied by the segment diameter.

                        % Plosive
                        plosiveDuration = round(nSamples/8);
                        plosiveEnv = linspace(0,1,plosiveDuration);
                        plosiveBefore = ones(1,(PlosiveStartPosition));
                        plosiveAfter = ones(1,(nSamples-PlosiveStartPosition-plosiveDuration));
                        plosiveLine = [plosiveBefore plosiveEnv plosiveAfter];
                        ...

                        ...
                        if plosive
                            plo = plosiveLine(n);
                            A(44) = plo*A(44);
                        end
                        

The table has audio examples of plosives at different positions, the excitation type was “LF” and the vowel synthesised was ε (bird).

After creating plosives and excitation, the next feature to look at was creating diphthongs. These are created by transitioning from one monophthong to another mid-way through. The code allows the user to specify two monophthongs, and transition between the two. This allows the user to create sounds that might not be normally produced.

                    % Dipthong transitions
                    transition = linspace(0,1,nSamples);
                    transition2 = 1 - transition;
                    ...

                    ...
                    if dipthong
                        % for when we're using dipthongs we need to move from one vowel to
                        % the next over time
                        for j=1:nSegments
                            A(j) = transition2(n)*firstVowel(j) + transition(n)*secondVowel(j);
                        end
                    else
                        % Select a vowel for synthesis
                        for j=1:nSegments
                            A(j) = firstVowel(j);
                        end
                    end
                    

Audio Examples

Analysis

Evaluation and Further Development

In the future it would be good to extend the model to produce more phonemes and possibly even combine voices to create a “virtual choir”. The model could also be developed further adding additional boundary conditions and expanding to use different LF excitement or excitement types in combination. It could also be improved by increasing the spatial sampling resolution which would improve the simulation of formants and create a more realistic vowel synthesis. The model could also be expanded to 2D or even 3D. Moving to more dimensions allows mouth features to be modelled and a much more natural sounding model.

The implemented extension of the model was to add diphthongs and plosives, however the user can also explore different excitation types and change the start position of plosives. The model created provided a strong starting point to model the vocal tract creating a reasonably accurate outcome.

Word Count : 1625

References