Linear Predictive Coding is a speech analysis technique used to estimate the basic parameters of speech as pitch (through frequency) and intensity (through loudness). Here, it is implemented and studied through Graphical User Interface(GUI) in MATLAB.
Linear Predictive Coding is a speech analysis technique used to estimate the basic parameters of speech. Speech samples can be approximated as a linear combination of the past samples by minimizing the error. This is done by minimizing the sum of squared differences between the actual speech signal samples and the linearly predicted ones. The sampled input speech signal is applied to an analyzer, which determines the parameters of the speech signal to be transmitted to the synthesizer. This speech synthesizer reconstructs the approximated speech signal. The input provided to the encoder is the comparison between the sampled signal and the approximated signal with the parameters. This encoder forms the digital signal known as LPC as output. This LPC output is provided to the Low Pass Filter, which reconstructs the audio signal x(t) by performing the interpolation of samples in the input.
- Graphical User Interface(GUI) is created in MATLAB using Guide in command window.
- It has a button for recording the speech which records the sound for 5 seconds and is saved as a file and a play button which plays the audio and plots the graph of the input speech signal.
- It takes the input length of audio segment in ms and the percentage of overlap from the user which segments it down into smaller segments.
- It has a dropdown menu which asks user to select the type of window(Hanning, Hamming, Bartlett, Blackman).
- It has a dropdown menu which asks use to slect the order of the filter(12,48,72,96)
- The sampled input speech signal is fed to analyzer for calculating the the filter coefficients of LP Filter and the pitch of each segment which is then fed to syntesizer for further operations.
- It has a button for pitch calculation which on clicking plots the pitch of each segment of the input speech signal.
- It has a button for reconstruction of signal without pitch and reconstruction of the signal with pitch. On clicking it shows the output graph and reconstructs the signal.
- The approximated speech signal is then reconstructed by syntesizer. The input to the encoder is the comparison between the sampled signal and the approximated signal with the parameters.
- The output obtained from encoder is the LPC output.
- The LPC output is then passed to to the Low Pass Filter for reconstruction of the speech signal using interpolation of samples of the input with two cases that are, with and without pitch information.
The quality of the reconstructed speech signal output was relatively low. As compared to the original speech signal, due to higher rate of compression, the output speech signal was distorted and less legible. The pitch of output signal was also low, as the output was deeper than the input speech.
The quality of output speech signal increases as the number of previous samples (order of the LP filter) for prediction increases. Thus, due to a lesser rate of compression (higher order filter – 48), the reconstructed speech signal was less distorted and more legible. The pitch of output was also marginally higher than before.
As order of the LP filter increases, the quality of output speech signal increases. The speech is much more understandable due to very less distortion. The distortion is less as the rate of compression is less. Thus, overall output quality is better for the LP filter of order 72.
Quality of output speech signal is the best for a LP filter of order 96. Since 96 previous samples have been considered to reconstruct the output, the compression rate is less, and thus the distortion of the output is also less. The output is completely legible and quality is not coarse, as compared to the previous lower order filters.
| Order of LP Filter | Without Pitch Detection | With Pitch Detection |
|---|---|---|
| 12 | 9.3 to 1 | 8.6 to 1 |
| 48 | 2.45 to 1 | 2.35 to 1 |
| 72 | 1.65 to 1 | 1.57 to 1 |
| 96 | 1.24 to 1 | 1.20 to 1 |
- With Pitch Detection:
- The rate of compression of input speech signal is reduced and hence the distortion in the output is also reduced.
- It produces a clear output and the nasality tone of the output was more prominent.
- Without Pitch Detection:
- It produces a signal with higher pitch/depth.
click here to view the demonstration.