### Speech spectral reduction for robust automatic speech recognition

In purpose of improving automatic speech recognition rate on embedded device, we've decided to use the FAME strategy known as "Frequency and Amplitude Modulation Encoding" proposed by Wei-Dong Wang, Hong-Yun Liu, Hu Yuan and Qing Ang in their article **A New Speech Coding Strategy for Cochlear Implant** which is based on the fact that speech signal can be represented as a sum of signals in given subbands and is expressed by the following equation.

$$y(t) = \sum_{k=0}^{N}{m_k(t) * \cos[2\pi\textit{f}_{c_k}(t) + \phi(t)]} $$

Where:

$$y(t)$$ |
: Represent the synthesized signal. |

$$N$$ |
: The number of subbands |

$$m_k$$ |
: The \(k^{th}\) band amplitude signal that modulate the carrier signal. |

$$\textit{f}_{c_k}(t)$$ |
: The \(k^{th}\) band central frequency. |

$$\phi(t)$$ |
: The signal phase |

we can summurise the FAME strategy with the following graphic

## Follow us