SINUSOIDAL TRANSFORM CODING
<b>ABSTRACT:/b><br> <br /> In digital speech compression, a Sinusoidal Transform Coder has been successfully applied on speech and music. This method uses sinusoids to model the excitation rather than the classical method based on pulses. In this thesis, we employ this sinusoid...
Saved in:
| Main Author: | |
|---|---|
| Format: | Theses |
| Language: | Indonesia |
| Online Access: | https://digilib.itb.ac.id/gdl/view/3088 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Institut Teknologi Bandung |
| Language: | Indonesia |
| Summary: | <b>ABSTRACT:/b><br> <br />
In digital speech compression, a Sinusoidal Transform Coder has been successfully applied on speech and music. This method uses sinusoids to model the excitation rather than the classical method based on pulses. In this thesis, we employ this sinusoidal model for application at low bit-rate speech coding. The sinusoidal model requires a good model of the spectrum, fundamental frequency and voicing decisions. In this thesis, a linear interpolation method is used to interpolate spectrum's peaks. A cepstrum method is used to model the spectrum envelope.</p> <br />
<br />
The synthesized speech is done by the way of modelling signal spectrum. The signal spectrum is sampled at a multiply of pitch for voiced condition. For unvoiced, speech is modelled with a fixed pitch frequency of 100 Hz. All informations from sampled spectrum is used to generate sinusoids and add together.</p> <br />
<br />
In this thesis, cep strum method is used to model the STFT (Short-Term Fourier Transform) spectrum envelope and only 20 coefficients are transmitted. Before transmitted, all parameters are quantized using uniform quantization. Frame duration of 160 samples (20 milli seconds) is used. The coder in this thesis has frame rate of 50 frames per second and could achieve bit rate of 8800 bit per second. The coder has average segSNR of 2.0909 dB. |
|---|