Transformations of LPC and LSF Parameters to Speech Recognition Features

LSF and LPC - derived features for Large Vocabulary Distributed Continuous Speech Recognition in Brazilian Portuguese

“…In this paper, we will consider only the MEL scale features (MLPCC, MPCC and MPCEP), since they provide a much better performance than the ones achieved with the linear scale features (LPCC, PCC and PCEP) [2]. The MFCC (Mel-Frequency Cepstral Coefficients) features [8]- [9] will be also obtained from voice reconstructed with the ITU-T G.723.1 codec at the two different operation rates (6.3 kbit/s or 5.3 kbit/s) [4].…”

“…Since the purpose of each filtering operation is to approximate the mel scale frequency, it is important to analyze the relationship of the transfer function given by (2) and the transformation of the frequency axis. In order to simplify the filter implementation, let a be a real number [12].…”

“…However, speech coders for mobile telephony and IP networks generate their parameters at lower rates (e.g., 50 Hz or 33 Hz). In a recent study on the efficiency of recognition features for distributed speech recognition [2], it was shown that low rates significantly degrade the performance of the recognizer. Hence, it is paramount to interpolate the speech features in order to achieve a recognition performance which is closer to the one obtained when the features are extracted at a high rate.…”

Recuperação de Pacotes Perdidos em Sistemas de Reconhecimento de Voz Distribuído usando Redes Neurais

“…reconhecimento. Neste artigo, nós consideraremos somente o atributo MPCEP (Mel Frequency Pseudo Cepstrum), pois o mesmo demonstrou em [4] que fornece um desempenho melhor, com uma complexidade menor, do que outros atributos de reconhecimento obtidos dos parâmetros do codec.…”