WORLD: A Vocoder-Based High-Quality Speech Synthesis System for Real-Time Applications

By Masanori Morise, Fumiya Yokomori, Kenji Ozawa. 2016

1,203 citations recorded by the source corpus.

SUMMARY A vocoder-based speech synthesis system, named WORLD, was developed in an e ﬀ ort to improve the sound quality of real-time applications using speech. Speech analysis, manipulation, and synthesis on the basis of vocoders are used in various kinds of speech research. Although several high-quality speech synthesis systems have been developed, real-time processing has been di ﬃ cult with them because of their high computational costs. This new speech synthesis system has not only sound quality but also quick processing. It consists of three analysis algorithms and one synthesis algorithm proposed in our previous research. The e ﬀ ectiveness of the system was evaluated by comparing its output with against natural speech including consonants. Its processing speed was also compared with those of conventional systems. The results showed that WORLD was superior to the other systems in terms of both sound quality and processing speed. In particular, it was over ten times faster than the conventional systems, and the real time factor (RTF) indicated that it was fast enough for real-time processing.

Subfields: Artificial Intelligence, Natural Language Processing, Signal Processing

How researchers see this paper

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

In Artificial Intelligence

In Natural Language Processing

In Signal Processing

Citations in the corpus

WaveNet: A Generative Model for Raw Audio (2016, background): “…a number of attempts to address these issues individually, such as developing high-quality vocoders (Kawahara et al., 1999; Agiomyrgiannakis, 2015; Morise et al., 2016), improving the accuracy of generative models (Zen et al., 2007; 2013; Fan et al., 2014; Uria et al., 2015), and compensating the…”
Deep Voice: Real-time Neural Text-to-Speech (2017, methodology): “Char2Wav relies on vocoder features from the WORLD TTS system (Morise et al., 2016) for pre-training their alignment module which include F0, spectral envelope, and aperiodic parameters.”
Vocos: Closing the gap between time-domain and Fourier-based neural vocoders for high-quality audio synthesis (2023, background): “While vocoders evolved significantly (Kawahara et al., 1999; Morise et al., 2016), they tended to oversimplify speech production, generating a distinctive ”buzzy” sound and thus compromising the naturalness of the speech.”
Audiodec: An Open-Source Streaming High-Fidelity Neural Audio Codec (2023, positive): “The WORLD vocoder [38] was utilized to extract F 0 , U/V flags, and mel-cepstral coefficients ( mcep ).”
A Survey on Neural Speech Synthesis (2021, methodology): “In statistical parametric speech synthesis [409, 350, 408, 418, 351], LSP (line spectral pairs) [132], MCC (mel-cepstral coefficients) [79], MGC (mel-generalized coefficients) [349], F0 and BAP (band aperiodicities) [153, 154] are used as acoustic features, which can be easily converted into wavefor”
TTS-Portuguese Corpus: a corpus for speech synthesis in Brazilian Portuguese (2020, methodology): “[18] proposed a fully convolutional model for speech synthesis and compared three different vocoders: Griffin-Lim [28], WORLD Vocoder [29] and WaveNet [30].”
Sample Efficient Adaptive Text-to-Speech (2018, positive): “, 2018) models produce World Vocoder Features (Morise et al., 2016) instead of generating raw audio signals.”
The Emotional Voices Database: Towards Controlling the Emotion Dimension in Voice Generation Systems (2018, methodology): “The VC module do this by extracting speech features with a vocoder (the default WORLD vocoder [18] was used) of both source and target voices, performing a Dynamic Time Warping (DTW) to align the features in time and computing a regression between the source and target features.”
Fitting New Speakers Based on a Short Untranscribed Sample (2018, background): “Char2Wav and Voice Loop produce World Vocoder Features (Morise et al., 2016).”
Deep Voice 3: 2000-Speaker Neural Text-to-Speech (2017, methodology): “Char2Wav predicts the parameters of WORLD vocoder (Morise et al., 2016) and uses a Sam-pleRNN conditioned upon WORLD parameters for waveform generation.”
ClariNet: Parallel Wave Generation in End-to-End Text-to-Speech (2018, methodology): “However, they depend on a traditional vocoder (Morise et al., 2016), the Grifﬁn-Lim algorithm (Grifﬁn and Lim, 1984), or a separately trained neural vocoder (Ping et al., 2018; Shen et al., 2018) to convert the predicted spectrogram to raw audio.”
DDSP: Differentiable Digital Signal Processing (2020, background): “Known as vocoders or synthesizers, these models are physically and perceptually motivated and have a long history of research and applications (Beauchamp, 2007; Morise et al., 2016).”
MelGAN: Generative Adversarial Networks for Conditional Waveform Synthesis (2019, methodology): “For instance, the WORLD vocoder (MORISE et al., 2016) introduces an intermediate representation tailored to speech modelling based on mel-spectrogram-like features.”
Adversarial Audio Synthesis (2018, methodology): “These vocoder features are synthesized to raw audio using off-the-shelf methods such as WORLD (Morise et al., 2016) and Grifﬁn-Lim (Grifﬁn & Lim, 1984), or trained neural vocoders (Sotelo et al., 2017; Shen et al., 2018; Ping et al., 2018).”
DiffSinger: Singing Voice Synthesis via Shallow Diffusion Mechanism (2021, methodology): “1; 3) FFT-NPSS [2] (WORLD), the SVS system which generates WORLD vocoder features [19] through feed-forward Transformer (FFT) and uses WORLD vocoder to synthesize audio; 4) FFT-Singer (Mel + PWG) the SVS system which generates mel-spectrograms through FFT network and uses PWG vocoder to synthesize a”
Neural Source-Filter Waveform Models for Statistical Parametric Speech Synthesis (2019, methodology): “Two types of spectral features were prepared: Mel-generalized cepstral coefficients (MGC) [48] of order 60 extracted by using the WORLD vocoder [49] and Mel-spectrogram of order 80.”
Voice Synthesis for in-the-Wild Speakers via a Phonological Loop (2017, methodology): “We, therefore, do not train an end-to-end system and employ the WORLD [16] vocoder (D4C edition [15]) for feature extraction.”
Neural Source-filter-based Waveform Model for Statistical Parametric Speech Synthesis (2018, methodology): “Acoustic features, including 60 dimensions of Mel-generalized cepstral coefﬁcients (MGCs) [18] and 1 dimension of F0, were extracted from the 48 kHz waveforms at a frame shift of 5 ms using WORLD [19].”