BestHub
search
Sign in
Tag

neural TTS

1 views collected around this technical thread.

Xiaohongshu Tech REDtech
Xiaohongshu Tech REDtech
Aug 10, 2022 · Artificial Intelligence

Multi-Stage Multi-Codebook VQ-VAE for High-Performance Neural Text-to-Speech (MSMC‑TTS)

The MSMC‑TTS system, a multi‑stage multi‑codebook VQ‑VAE based neural text‑to‑speech solution, delivers near‑human audio quality (MOS 4.41) with a compact 3.12 MB acoustic model, substantially surpassing Mel‑Spectrogram FastSpeech baselines in naturalness and efficiency.

Compact RepresentationMulti-Stage ModelingVQ-VAE
0 likes · 10 min read
Multi-Stage Multi-Codebook VQ-VAE for High-Performance Neural Text-to-Speech (MSMC‑TTS)
Zuoyebang Tech Team
Zuoyebang Tech Team
May 19, 2022 · Artificial Intelligence

How to Achieve High‑Quality TTS with Only Minutes of Data

This article reviews neural speech synthesis, explains why large high‑quality paired data are essential, and presents a range of low‑resource solutions—including semi‑supervised pre‑training, cross‑language transfer, speaker embedding, and Conformer‑based model upgrades—demonstrating how the Zuoyebang team built a robust TTS system with as little as 7‑minute speaker recordings.

Fastspeech2conformerlow-resource TTS
0 likes · 15 min read
How to Achieve High‑Quality TTS with Only Minutes of Data