# Neural source-filter-based waveform model for statistical parametric speech synthesis

arXiv: https://arxiv.org/abs/1810.11946 · HTML: https://mlsanity.com/paper/1810.11946/

## Key facts

- Cited by 22 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 29 Oct 2018.

## Built on by (most cited first, 22 of 22)

- [DiffWave: A Versatile Diffusion Model for Audio Synthesis](https://mlsanity.com/paper/2009.09761/) (2020)
- [A Survey on Neural Speech Synthesis](https://mlsanity.com/paper/2106.15561/) (2021)
- [Neural Analysis and Synthesis: Reconstructing Speech from Self-Supervised Representations](https://mlsanity.com/paper/2110.14513/) (2021)
- [The Ethical Implications of Generative Audio Models: A Systematic Literature Review](https://mlsanity.com/paper/2307.05527/) (2023)
- [Real-time Denoising and Dereverberation with Tiny Recurrent U-Net](https://mlsanity.com/paper/2102.03207/) (2021)
- [FastPitchFormant: Source-filter based Decomposed Modeling for Speech Synthesis](https://mlsanity.com/paper/2106.15123/) (2021)
- [FSD: An Initial Chinese Dataset for Fake Song Detection](https://mlsanity.com/paper/2309.02232/) (2023)
- [Source-Filter HiFi-GAN: Fast and Pitch Controllable High-Fidelity Neural Vocoder](https://mlsanity.com/paper/2210.15533/) (2022)
- [Towards single integrated spoofing-aware speaker verification embeddings](https://mlsanity.com/paper/2305.19051/) (2023)
- [A Benchmark for Multi-speaker Anonymization](https://mlsanity.com/paper/2407.05608/) (2024)
- [Period VITS: Variational Inference with Explicit Pitch Modeling for End-to-end Emotional Speech Synthesis](https://mlsanity.com/paper/2210.15964/) (2022)
- [Unified Source-Filter GAN with Harmonic-plus-Noise Source Excitation Generation](https://mlsanity.com/paper/2205.06053/) (2022)
- [Unsupervised Music Source Separation Using Differentiable Parametric Source Models](https://mlsanity.com/paper/2201.09592/) (2022)
- [Hear Your Face: Face-based voice conversion with F0 estimation](https://mlsanity.com/paper/2408.09802/) (2024)
- [Unified Source-Filter GAN: Unified Source-filter Network Based On Factorization of Quasi-Periodic Parallel WaveGAN](https://mlsanity.com/paper/2104.04668/) (2021)
- [High-Fidelity and Low-Latency Universal Neural Vocoder based on Multiband WaveRNN with Data-Driven Linear Prediction for Discrete Waveform Modeling](https://mlsanity.com/paper/2105.09856/) (2021)
- [SPA-SVC: Self-supervised Pitch Augmentation for Singing Voice Conversion](https://mlsanity.com/paper/2406.05692/) (2024)
- [Real-time Timbre Remapping with Differentiable DSP](https://mlsanity.com/paper/2407.04547/) (2024)
- [HQ-SVC: Towards High-Quality Zero-Shot Singing Voice Conversion in Low-Resource Scenarios](https://mlsanity.com/paper/2511.08496/) (2025)
- [Low-Latency Real-Time Non-Parallel Voice Conversion based on Cyclic Variational Autoencoder and Multiband WaveRNN with Data-Driven Linear Prediction](https://mlsanity.com/paper/2105.09858/) (2021)
- [Improving Inference-Time Optimisation for Vocal Effects Style Transfer with a Gaussian Prior](https://mlsanity.com/paper/2505.11315/) (2025)
- [Paradee: Distilling Kokoro-82M into an 8M-Parameter Single-Voice Text-to-Speech Model](https://mlsanity.com/paper/2610.06817/) (2026)

## Abstract (arXiv)

Neural waveform models such as the WaveNet are used in many recent text-to-speech systems, but the original WaveNet is quite slow in waveform generation because of its autoregressive (AR) structure. Although faster non-AR models were recently reported, they may be prohibitively complicated due to the use of a distilling training method and the blend of other disparate training criteria. This study proposes a non-AR neural source-filter waveform model that can be directly trained using spectrum-based training criteria and the stochastic gradient descent method. Given the input acoustic features, the proposed model first uses a source module to generate a sine-based excitation signal and then uses a filter module to transform the excitation signal into the output speech waveform. Our experiments demonstrated that the proposed model generated waveforms at least 100 times faster than the AR WaveNet and the quality of its synthetic speech is close to that of speech generated by the AR WaveNet. Ablation test results showed that both the sine-wave excitation signal and the spectrum-based training criteria were essential to the performance of the proposed model.