Performance Analysis of RECONSTRUCTION OF VOICED SPEECH SIGNALS
Task 1: Pitch Estimation in Time Domain (L1)
Step 1: Initially we collect recorded audio, pre-process the audio and perform frame division.
Step 2: Then we perform pitch estimation using Autocorrelation Function (AF) and Average Magnitude Difference Function (AMDF).
Task 2: Pitch Estimation in Frequency Domain (L2)
Step 1: Initially we collect recorded audio, pre-process the audio and perform frame blocking.
Step 2: Then we compute the Fast Fourier Transform (FFT) and the magnitude spectrum of each frame.
Step 3: Then we perform pitch estimation using Harmonic Product Spectrum (HPS) and Harmonic Sum Spectrum (HSS)
Task 3: Voiced and Unvoiced Frame Separation (L2 – L3)
Step 1: Initially we record an audio worded ‘six’ and implement a function to analyse the voiced and unvoiced segments using frame energy and pitch estimates.
Step 2: Next we create a tone for each pitch frequency and generate random noise for the unvoiced segments.
Task 4: Pitch Recovery from Telephone Bandwidth (L3)
Step 1: Initially we record a fully voiced sentence like “We were away”, Then design the band-pass filter and apply it to the recorded speech.
Step 2: then we implement a nonlinear circuit, again applying a band-pass filter and estimate the pitch period.
Software Requirement:
———————————–
Development tool: Matlab 2020a or above version.
Operating system: Windows 10 [64-Bits]
Note:
———–
1) If the above plan does not satisfy your requirement, please provide the processing details, like the above step-by-step.
2) Please note that this implementation plan does not include any further steps after it is put into implementation.
