Objective

Linear prediction is a method used to estimate a time-varying filter, often as a model of a vocal tract. Musical applications of linear prediction substitute various signals as excitation sources for the time-varying filter.

This mini-project will give you chance to develop the basic technique for computing and applying a time-varying filter. Next, you will experiment with different excitation sources and linear prediction model parameters. Finally, you will learn about cross-synthesis.

Prerequisite Modules

If you have not done so already, please study the prerequisite modules Linear Prediction and Cross Synthesis. If you are relatively new to LabVIEW, consider taking the course LabVIEW Techniques for Audio Signal Processing which provides the foundation you need to complete this mini-project activity, including working with arrays, creating subVIs, playing an array to the soundcard, and saving an array as a .wav sound file.

Deliverables

Part 1: Framing and De-Framing

Time-varying filters operate by applying a fixed set of coefficients on short blocks (or "frames") of the signal; the coefficients are varied from one frame to the next. In this part you will develop the basic technique used to "frame" and "de-frame" a signal so that a filter can be applied individually to each frame.

Download and open framing.vi.

The "Reshape Array" node forms the heart of framing and de-framing, since you can reshape the incoming 1-D signal vector into a 2-D array of frames. The auto-indexing feature of the "for loop" structure automatically loops over all of the frames, so it is not necessary to wire a value to the loop termination terminal. You can access the individual frame as a 1-D vector inside the loop structure. Auto-indexing is also used on the loop output to create a new 2-D array, so "Reshape Array" is again used to convert the signal back to a 1-D vector.

Study the entire VI, including the unconnected blocks which you will find useful. Complete the VI so that you can select frame sizes of between 1 and 9. Enable the "Highlight Execution" option, and display your block diagram and front panel simultaneously (press Ctrl-T). Convince yourself that your technique works properly. For example, when you select a frame size of 2, you should observe that the front-panel indicator "frame" displays "0,1", then "2,3", then "4,5", and so on. You should also observe that the "out" indicator matches the original.

Part 2: Time-Varying Filter Using Linear Prediction

Download the file part2.zip, a .zip archive that contains three VIs: part2.vi, blp.vi (band-limited pulse source), and WavRead.vi (reads a .wav audio file). Complete this VI by creating your own "Framer" and "DeFramer" VIs using the techniques you developed in Part 1.

Create or find a speech-type .wav file to use as a basis for the linear prediction filter. Vary the frame size and filter order parameters as well as the various type of excitation sources. Study the effect of each parameter and discuss your results. Submit one or two representative .wav files.

Part 3: Cross Synthesis

"Cross synthesis" applies the spectral envelope of one signal (e.g., speech) to another signal (e.g., a musical instrument). Find or create a speech signal and use it to generate a time-varying filter. Find or create a music signal and use it as the excitation. The sound files should have the same sampling frequency.

Repeat for a second set of signals. You might also try cross synthesizing two different speech signals or two different music signals.

Show your results, particularly the spectrograms of the two original signals and the spectrogram of the output signal.

Select your favorite result and submit .wav files of the two source signals and the output signal.