A Speech Enhancement System Based on Linear Prediction

In a tunnel on an express-way, emargency telephone boxes are settled to keep speech quality of a near-end talker from ambient noises.

When the talker opens the door to confirm an accident situation, the telephone box cannot reduce the noise.

We have developed a real-time noise suppressor for high-level noises in a tunnel.

Speech and noise waveforms are shown in the figure.

We can predict a present speech from past speech samples. But, noise prediction is difficult.

We employ a linear predictor which gives only predictable signals.


We implement the linear predictor onto a DSP (Digital Signal Processor).

Noise reduction results show that the proposed real-time noise suppressor is applicable fora high-level tunnel noise environment.



Waveform of noise reduction results


Spectrogram of noise reduction results

Sound Data
  • Observe Signal
  • Output of Conventional Method
  • Output of Proposed Noise Method

  • Back