Hello there, I’m a first-year PhD student in @aim-qmul at @c4dm, Queen Mary University of London, working on controllable and expressive neural voice synthesis 🎶

I'm interested in signal processing, music information retrieval, binaural audio, machine learning, or any audio-related tech.

PhD Projects

• diffwave-sr: unsupervised speech super-resolution (bandwidth extension) using posterior sampling in diffusion models.
• golf: a light-weight neural vocoder with glottal-flow models and differentiable LPC synthesis.

Tools

• torch_specinv: a collections of spectrogram inversion algorithms.
• torchnmf: a package that can help build complex NMF models.
• kazane: simple sinc interpolation for 1D signal in PyTorch.
• torch-fftconv: FFT-based PyTorch convolution operators.

Re-implementations

• wavenet-like-vocoder: WaveNet and FFTNet re-implementations.
• constant-memory-waveglow: training waveglow with constant memory cost.
• variational-diffwave: training DiffWave with unbiased ELBO.

I’m also the main contributor to makingtorchaudio lfilter differentiable. Your sponsorship will support my development and maintenance of the above tools, my engagement in open-source science, and my PhD research.

Current sponsors1

Past sponsors5

Private Sponsor

kweiwen sponsors this goal