Perceptual linear predictive (PLP) analysis of speech.

By Hynek Hermansky. 1990

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

1. Li Deng
2. Dong Yu
3. Yoshua Bengio
4. Jinyu Li
5. Tara N. Sainath
6. Shinji Watanabe
7. Bo Li
8. Kyunghyun Cho

Citations in the corpus

• Time-Frequency Masking in the Complex Domain for Speech Dereverberation and Denoising (2017, neutral): “These features include amplitude modulation spectrogram (AMS) [23], relative spectral transform and perceptual linear prediction (RASTA-PLP) [14], [15], mel-frequency cepstral coefficients (MFCC), as well as their deltas.”
• Acoustic monitoring in terrestrial environments using microphone arrays: applications, technological considerations and prospectus (2011, methodology): “Cepstral coefficients are commonly used in human speech processing and recognition (Davis & Mermelstein 1980; Hermansky 1990; Milner 2002) and have recently been used in some wildlife studies (Soltis, Leong & Savage 2005b; Mazhar, Ura & Bahl 2007) by adapting them for the different hearing…”
• Survey of the State of the Art in Human Language Technology (2010, methodology): “Perceptual Linear Prediction (PLP) combines the LPC and filter-bank approaches by fitting an all-pole model to the set of energies (or, strictly, loudness levels) produced by a perceptually motivated filter bank, and then computing the cepstrum from the model parameters (Hermansky, 1990).”
• Deep Neural Networks for Acoustic Modeling in Speech Recognition (2012, methodology): “IEEE SIGNAL PROCESSING MAGAZINE [3] NOvEMbER 2012 GMMs have a number of advantages that make them suitable for modeling the probability distributions over vectors of input features that are associated with each state of an HMM.”
• Discovering hierarchical speech features using convolutional non-negative matrix factorization (2003, background): “Mel-frequency cepstral coefficients (MFCC) [4] and perceptual linear predictive coefficients (PLP) [5] form the basis of many ASR systems.”
• Acoustic Modelling from the Signal Domain Using CNNs (2016, methodology): “based on fixed filters, such as MFCC and PLP.”
• Automatic speech emotion recognition using modulation spectral features (2011, methodology): “PLP features In addition to MFCCs, perceptual linear predictive (PLP) coefficients (Hermansky, 1990) are also extracted from speech, serving as an alternative choice of short-term spectral features for comparison.”
• Recognising realistic emotions and affect in speech: State of the art and lessons learnt from the first challenge (2011, background): “Perceptual Linear Predictive (PLP) coefficients [89] and MFCCs are extremely similar, as they both correspond to a short-term spectrum smoothing – the former by an autoregressive model, the latter by the cepstrum – and to an approximation of the auditory system by filter-bank-based methods.”
• Deep neural network features and semi-supervised training for low resource speech recognition (2013, methodology): “Separate DNNs are trained on two different feature representations, short-term spectral PLP features [30] and long-term FDLP-based modulation features [31].”
• Size matters: an empirical study of neural network training for large vocabulary continuous speech recognition (1999, methodology): “PLP-12 features [4], including the PLP gain term to give a 13-element feature vector, were computed every 16 ms, and normalized according to the mean and variance of each segment in the training data.”
• The importance of phase in speech enhancement (2011, background): “For example, the automatic speech and speaker recognition systems derive cepstral features from the magnitude spectrum alone (e.g., Davis and Mermelstein, 1980; Hermansky, 1990).”
• Connectionist probability estimators in HMM speech recognition (1994, methodology): “Using single frame perceptual linear prediction (PLP) coefficient inputs [47], a 512 hidden unit MLP produced a phone classification rate of 59%, at the frame level.”
• Robust speech recognition using the modulation spectrogram (1998, methodology): “The two RASTA-PLP algorithms are temporal processing extensions to perceptual linear prediction (PLP) (Hermansky et al., 1985; Hermansky, 1990), an algorithm for producing a spectral shape estimate that re ̄ects properties of the human auditory system.”
• Large Vocabulary Continuous Speech Recognition : a ReviewSteve (1996, methodology): “Good results have also been reported using LP coe cients to derive a smoothed spectrum which is then perceptually weighted to give Perceptually weighted Linear Prediction (PLP) coe cients[44].”