Advanced editing tools

Snoori offers the following tools to edit signals:

Actions accessible through the mouse depend on the modification selected. The bottom right radio button list allows the type of modification to be selected. The default pointing mode is “select” and corresponds to the selection of a region with the pointer.

Changing the modification highlights the sub window that allows editing parameters to be adjusted.

Note that for all these modifications the Apply button must be pushed to start modifications.

The Cancel button enables only the last modification to be cancelled.

Damping (left, middle or right) enables the attenuation the left, the middle or the right portion of the highlighted region.

  1. The attenuation is said to be “left” when the multiplying factor starts at 0 at the left end and increases linearly to 1 towards the right end of the highlighted region.

  2. The attenuation is said to be “middle” when the speech signal of the highlighted region is multiplied by a coefficient the gain (in dB) of which is to be specified in the panel. The coefficient decreases, or increases, progressively from 1 to the user specified value over a portion extending over 20 milliseconds but not exceeding the third of the region.

  3. This attenuation is said to be “right” when the attenuation profile is the symmetrical of that of the left attenuation (the multiplying factor starts at 1 at the left end and decreases linearly to 1 towards the right end of the highlighted region).

FIR filtering enables the definition of a FIR filter applied to a given time x frequency region. Order, attenuation (in dB) and type of filter (stop or pass band) can be chosen. Once these parameters are set, the user can draw a rectangle with the pointer to select the time x frequency region where to apply the filtering.

Adding noise enables noise to be added to a signal. Before setting the Signal to Noise Ratio (SNR) the user has to compute the level of the speech signal. If a signal region is selected Snoori computes the level of speech over this region. Otherwise, the level is computed over the signal in the memory of Snoori. Note that this preliminary calculation has a strong influence on the noise that is added to signal. If the signal is zero it receives a default level of 60 dB. Note that the default spectrogram calculation uses pre-emphasis and therefore gives the feeling that high frequencies are more affected than low frequencies.

OLA filtering enables arbitrary time x frequency trajectories to be filtered by the overlap and add method. User has to draw trajectories. Then, trajectories can be edited, saved, read and applied to a signal. For each of these trajectories, bandwidth and gain can be specified independently.

The upper parameters and actions (i.e. default width and gain, stop or pass, read, save, apply and forget) concern all the trajectories whereas bottom parameters and actions concern selected trajectories only.

Mouse actions are quite similar to those of the Klatt synthesizer mode, i.e.:

The OLA overlap is 75%, FFT order is 1024 and the signal window (32 ms) is zero padded.

OLA filtering in Snoori was designed with a view of lowering or raising formants or harmonics levels. Therefore, the two commands “lock on LPC” and “lock on harmonics” can be used to move a trajectory towards the nearest line of LPC roots or narrow band spectrum peaks.

Note that it is by far much more convenient to compute a wide band spectrogram before drawing trajectories that should correspond to formants and a narrow ban spectrogram before drawing trajectories supposed to correspond to harmonics.