A new mixed excitation LPC vocoder

Speech coding

“…Such schemes not only led to elimination of excitation sequences that were purely based on pitch estimation and voiced/unvoiced decisions, but also consequently led to reduction of some of the buzzy artifacts in voiced segments of reconstructed speech signal due to excessive periodicity [ 16,17,18,19,20]. Among these two schemes, namely Residual Excited Linear Predictive (RELP) coder and Mixed Excited Linear Predictive (MELP) coders gained prominence in this effort which are described briefly below.…”

“…For speech coders that are required to operate at bit rates of 4.8 kbps and below, RELP coders render themselves unsuitable. The Mixed Excitation Linear Predictive coders [17,18] provides a low bit rate alternative to RELP, and at the same time removes the inadequacies of the two-state excitation modeling, both in terms of removing buzziness in reconstructed speech as well as providing robustness to voicing decision errors. Unlike RELP coders, where pitch estimation was absent, MELP coders perform pitch estimation and incorporates frequency dependent voicing measures to model the LPC residual signal.…”

1kb/s waveform interpolation speech coding based on non-negative matrix factorization

“…At present, several existed speech coders with bit rate between 2 and 4kb/s have given an acceptable quality for communication applications, such as mixed excitation linear prediction (MELP) [3]. Several speech coders at very low bit rates have been proposed based on MELP algorithm by exploiting both intra-and interframe correlation of the parameters, i.e.…”

Avaliação Subjetiva de Versão Acelerada do Codificador de Voz MELP

“…Sua versão mais moderna oferece também as opções de 1200 e 600 bps, obtidas basicamente por meio de quantização vetorial de blocos de parâmetros. O grande mérito do MELP foi superar as limitações do antigo codificador de baixa taxa LPC10 (estabelecido na norma americana Federal Standard 1015), que também opera a 2400 bps, conferindo melhor qualidade de audio e maior robustez ao ruído ambiente [4] [5]. Essa melhoria de qualidade foi possível graças ao uso de um modelo mais sofisticado de produção da voz, ao custo de uma maior complexidade computacional.…”