Line spectrum representation of linear predictor coefficients of speech signals

By F. Itakura. 1975

Subfields: Evaluation & Benchmarks, Speech & Audio

How researchers see this paper

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

In Evaluation & Benchmarks

In Speech & Audio

Citations in the corpus

WaveNet: A Generative Model for Raw Audio (2016, background): “It often includes cepstra (Imai & Furuichi, 1988) or line spectral pairs (Itakura, 1975), which represent vocal tract transfer function, and fundamental frequency ( F 0 ) and aperiodicity (Kawahara et al., 2001), which represent characteristics of vocal source excitation signals.”
A Survey on Neural Speech Synthesis (2021, methodology): “Unlike the vocoder analysis and synthesis in SPSS [79, 349, 153, 132, 152, 232], WaveNet incorporates almost no prior knowledge about audio signals, and purely relies on end-to-end learning.”
Wavenet Based Low Rate Speech Coding (2017, background): “4 kb/s, each 20 ms block encodes the short-time spectral envelope using line spectral frequencies [20] with 36 bits, the pitch with 7 bits, the signal power with 5 bits, and the voicing level with 2 bits.”
Improving Opus Low Bit Rate Quality with Neural Speech Synthesis (2019, background): “The number of bits allocated to the filter (represented as line spectral pairs [21]) does not vary significantly with the total bit rate.”
An investigation of multi-speaker training for wavenet vocoder (2017, methodology): “The second approach, parametric approach, is a technique to generate a synthesized speech by utilizing acoustic features and vocoder frameworks which represent physical characteristics of human speech production [3, 4, 5, 6, 7, 8, 9, 10].”
Speech Synthesis Based on Hidden Markov Models (2013, background): “In addition to the mel-cepstral coefficients, various spectral representations, such as line-spectral pairs (LSPs) [43], mel-generalized cepstral coefficients [44], and various excitation parameters (e.”
A Deep Learning Approach to Data-driven Parameterizations for Statistical Parametric Speech Synthesis (2014, background): “Therefore, only representations like MCEPs or Line Spectral Pairs[28] have been found to be suitable.”
Directly modeling speech waveforms by neural networks for statistical parametric speech synthesis (2015, unknown): “…model for statistical parametric speech synthesis (SPSS) [1], a set of parametric representation of speech (e.g. cepstra [2], line spectrum pairs [3], fundamental frequency, and aperiodicity [4].) at every 5 ms is first extracted then relationships between linguistic features associated with…”
Minimum generation error training with direct log spectral distortion on LSPs for HMM-based speech synthesis (2008, methodology): “Although we used line spectral pairs (LSP) [10] as the spectral feature for HMM modeling, the Euclidean distance between two LSPs is not so convincing as a spectral distortion measure.”
Directly modeling voiced and unvoiced components in speech waveforms by neural networks (2016, background): “Index Terms— Statistical parametric speech synthesis; neural network; wavefom”
Super-Resolution from Noisy Data (2012, background): “In many of these applications, the signal we wish to super-resolve is a superposition of point sources; depending upon the situation, these may be celestial bodies in astronomy [26], molecules in microscopy [20], or line spectra in speech analysis [21].”
CommanderSong: A Systematic Approach for Practical Adversarial Voice Recognition (2018, methodology): “Common acoustic feature extraction algorithms include Mel-Frequency Cepstral Coefficients (MFCC) [40], Linear Predictive Coefficient (LPC) [34], Perceptual Linear Predictive (PLP) [30], etc.”
Techniques for Harmonic Sinusoidal Coding (1997, methodology): “An important property of the polynomials is that the roots appear in complex conjugate pairs on the unit circle [57].”