Cepstral analysis synthesis on the mel frequency scale

By Satoshi Imai. 1983

How researchers see this paper

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

Citations in the corpus

A Survey on Neural Speech Synthesis (2021, methodology): “The basic idea is that instead of direct generating waveform through concatenation, we can first generate the acoustic parameters [79, 349, 153] that are necessary to produce speech and then recover speech from the generated acoustic parameters using some algorithms [129, 128, 152, 232].”
Conv-TasNet: Surpassing Ideal Time-Frequency Magnitude Masking for Speech Separation (2018, methodology): “frequencies (more than 60% tuned to frequencies below 1 kHz). This pattern of frequency representation, which we found using a data-driven method, roughly resembles the well-known mel frequency scale [57] as well as the tonotopic organization of the frequencies in the mammalian auditory system [58],”
Automatic Building of Synthetic Voices from Audio Books (2010, methodology): “MCEP vectors are extracted as filter parameters and fundamental frequency (F0) estimates are derived as excitation features for every 5 ms [Imai, 1983].”
Spectral Mapping Using Artificial Neural Networks for Voice Conversion (2010, methodology): “Mel-cepstral coefficients (MCEPs) are extracted as filter parameters and fundamental frequency estimates are derived as excitation features for every 5 ms [22].”
An HMM-based singing voice synthesis system (2006, methodology): “Finally, a speech waveform is synthesized directly from the generated spectral and excitation parameters using Mel Log Spectrum Approximation (MLSA) ﬁlter [10].”
Speech acoustic modeling from raw multichannel waveforms (2015, methodology): “[11] improve performance by training a model on linear frequency FFf magnitude features to learn filterbank weights automatically rather than relying on the mel scale [12].”
Mel-generalized cepstral analysis - a unified approach to speech spectral estimation (1994, background): “The phase response βα(ω) gives a good approximation to auditory frequency scales with an appropriate choice of α [12], [19].”
COVAREP — A collaborative voice analysis repository for speech technologies (2014, background): “COVAREP includes an estimator of the so called “trueenvelope” (TE) [20, 21, 22].”
Statistical parametric speech synthesis: from HMM to LSTM-RNN (2015, methodology): “Estimate model parameters based on ML c = arg max c p(x | c) • p(x | c): AR model → Linear predictive analysis [5] • p(x | c): EX model → (ML-based) cepstral analysis [6]”
The HMM-based speech synthesis system (HTS) version 2.0 (2007, methodology): “Finally, a speech waveform is synthesized directly from the generated spectral and excitation parametersusing the correspondingspeech synthesisfilter (mel log spectrumapproximation(MLSA)filter [16] formel-cepstral coefficients).”
An overview of voice conversion systems (2017, methodology): “Commonly, mel-cepstrum is used in the literature ( Imai, 1983 ).”
Text-to-Speech Synthesis (2009, background): “A number of techniques have been developed which attempt to r everse these operations and generate speech [229], [261], [419], [85].”
Continuous Stochastic Feature Mapping Based on Trajectory HMMs (2011, methodology): “The speech waveform was resynthesized from the converted mel-cepstral coefficients and values through the mel log spectrum approximation (MLSA) filter [12] with pulse-train or white-noise excitation.”
Simultaneous Modeling of Spectrum, Pitch and Duration in HMM-Based Speech Synthesis (1999, methodology): “The mel-cepstral analysis technique enables speech to be re-synthesized from the mel-cepstral coefﬁcients using the MLSA (Mel Log Spectrum Approximation) ﬁlter [6].”
Phonetic posteriorgrams for many-to-one voice conversion without parallel data training (2016, background): “Mel-cepstral coefficients (MCEPs) [15] are extracted to represent the spectral envelope and then MCEPs features from the same sentences of the source speech and the target speech are aligned by dynamic time warping (DTW).”