Academia.eduAcademia.edu

Figure from:Speech Recognition: A Review

See full PDFdownloadDownload figure
International Journal of Advance Engineering and Research Development (IJAERD Volume 1,Issue 12, December -2014, e-ISSN: 2348 - 4470, print-ISSN:2348-640

Figure 2 International Journal of Advance Engineering and Research Development (IJAERD Volume 1,Issue 12, December -2014, e-ISSN: 2348 - 4470, print-ISSN:2348-640

Related Figures (5)

Audio input: the spoken speech is taken as an input to the system by means of microphone or any such device. Feature Extraction: this block does the recording of various speech samples. Digitization of signal takes place in this block where samp ling and quantization are performed. Elimination of noise in speech signal is done by quantization. Acoustic Modeling: the Acoustic modeling provides statistical model of speech signal. The acoustic model assigns probabilities to phonemes, words, or sentences. Language Modeling: Plain acoustic information is not enough knowledge about the language is also necessary. Comparison of symbols from acoustic model is done with the set of words present in the Dictionary to produce required sequence of words. Decoder is then used for recognizing the exact word.
Audio input: the spoken speech is taken as an input to the system by means of microphone or any such device. Feature Extraction: this block does the recording of various speech samples. Digitization of signal takes place in this block where samp ling and quantization are performed. Elimination of noise in speech signal is done by quantization. Acoustic Modeling: the Acoustic modeling provides statistical model of speech signal. The acoustic model assigns probabilities to phonemes, words, or sentences. Language Modeling: Plain acoustic information is not enough knowledge about the language is also necessary. Comparison of symbols from acoustic model is done with the set of words present in the Dictionary to produce required sequence of words. Decoder is then used for recognizing the exact word.
LPC (Linear Predictive coding) analy zes the speech signal by estimating the formants, removing their effects fromthe speech signal, and estimating the intensity and frequency of the remaining buzz. The process of removing the formants is called inverse filtering, and the remaining signal is called the residue. In LPC system, each sample of the signal is expressed as a linear combination of the previous samples. This equation is called a linear predictor and hence it is called as linear predictive coding.The coefficients of the difference equation (the prediction coefficients) characterize the formants.LPC provides a good prediction algorithm but it is not well able to mimic the human auditory system.
LPC (Linear Predictive coding) analy zes the speech signal by estimating the formants, removing their effects fromthe speech signal, and estimating the intensity and frequency of the remaining buzz. The process of removing the formants is called inverse filtering, and the remaining signal is called the residue. In LPC system, each sample of the signal is expressed as a linear combination of the previous samples. This equation is called a linear predictor and hence it is called as linear predictive coding.The coefficients of the difference equation (the prediction coefficients) characterize the formants.LPC provides a good prediction algorithm but it is not well able to mimic the human auditory system.
3.2 Linear Prediction Coefficients (LPC): Mel Frequency Cepstral Coefficients (MFCC) is one of the most commonly used feature extraction method in speect recognition. The technique is called FFT based which means that feature vectors are extracted from the frequency spectra o: the windowed speech frames. The Mel frequency filter bank is a series of triangular bandpass filters. The filter bank is basec on a non-linear frequency scale called the mel-scale. A 1000 Hz tone is defined as having a pitch of 1000 mel. Below 100 Hz, the Mel scale is approximately linear to the linear frequency scale. Above the 1000 Hz reference point, the relationshit between Mel scale and the linear frequency scale is non-linear and approximate ly Logarithmic. One advantage of MFCC is that it is able to mimic Human Auditory System well. But it is very sensitive t noise and has no prediction algorithm.
3.2 Linear Prediction Coefficients (LPC): Mel Frequency Cepstral Coefficients (MFCC) is one of the most commonly used feature extraction method in speect recognition. The technique is called FFT based which means that feature vectors are extracted from the frequency spectra o: the windowed speech frames. The Mel frequency filter bank is a series of triangular bandpass filters. The filter bank is basec on a non-linear frequency scale called the mel-scale. A 1000 Hz tone is defined as having a pitch of 1000 mel. Below 100 Hz, the Mel scale is approximately linear to the linear frequency scale. Above the 1000 Hz reference point, the relationshit between Mel scale and the linear frequency scale is non-linear and approximate ly Logarithmic. One advantage of MFCC is that it is able to mimic Human Auditory System well. But it is very sensitive t noise and has no prediction algorithm.
The probabilistic framework used in ASR builds on astatistical model of the speech signal. The role of the acoustic model is to evaluate acoustic feature sequences and assign probabilities for different acoustic classes. Depending on the recognizer, the acoustic model assigns probabilities to phonemes, words, or full sentences. International Journal of Advance Engineering and Research Development (IJAERD Volume 1,Issue 12, December -2014, e-ISSN: 2348 - 4470, print-ISSN:2348-640
The probabilistic framework used in ASR builds on astatistical model of the speech signal. The role of the acoustic model is to evaluate acoustic feature sequences and assign probabilities for different acoustic classes. Depending on the recognizer, the acoustic model assigns probabilities to phonemes, words, or full sentences. International Journal of Advance Engineering and Research Development (IJAERD Volume 1,Issue 12, December -2014, e-ISSN: 2348 - 4470, print-ISSN:2348-640
International Journal of Advance Engineering and Research Development (IJAERL Volume 1,Issue 12, December -2014, e-ISSN: 2348 - 4470, print-ISSN:234 8-640 The performance of speech recognition systems is normally referred in terms of accuracy and speed. Accuracy is labeled with word error rate (WER), Whereas speed is labelled with the real time factor. Alternately accuracy can be labelled Single Word Error Rate (SW ER) and Command Success Rate (CSR)[21].Word Error Rate (WER)[20], is a common measurement of the performance of a speech recognition. Normally problems occurred in the performance of system due to mismatch of recognized word sequence with reference word sequence. The WER is derived fromthe Levenshtein distance [20],operating at the word level. Word error rate calculated as fc*s.rMmraiiegy
International Journal of Advance Engineering and Research Development (IJAERL Volume 1,Issue 12, December -2014, e-ISSN: 2348 - 4470, print-ISSN:234 8-640 The performance of speech recognition systems is normally referred in terms of accuracy and speed. Accuracy is labeled with word error rate (WER), Whereas speed is labelled with the real time factor. Alternately accuracy can be labelled Single Word Error Rate (SW ER) and Command Success Rate (CSR)[21].Word Error Rate (WER)[20], is a common measurement of the performance of a speech recognition. Normally problems occurred in the performance of system due to mismatch of recognized word sequence with reference word sequence. The WER is derived fromthe Levenshtein distance [20],operating at the word level. Word error rate calculated as fc*s.rMmraiiegy