English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Neural text-to-speech (TTS) generally consists of cascaded architecture withseparately optimized acoustic model and vocoder, or end-to-end architecturewith continuous mel-spectrograms or self-extracted speech frames as theintermediate representations to bridge acoustic model and vocoder, whichsuffers from two limitations: 1) the continuous acoustic frames are hard topredict with phoneme only, and acoustic information like duration or pitch isalso needed to solve the one-to-many problem, which is not easy to scale onlarge scale and noise datasets; 2) to achieve diverse speech output based oncontinuous speech features, complex VAE or flow-based models are usuallyrequired. In this paper, we propose FoundationTTS, a new speech synthesissystem with a neural audio codec for discrete speech token extraction andwaveform reconstruction and a large language model for discrete tokengeneration from linguistic (phoneme) tokens. Specifically, 1) we propose ahierarchical codec network based on vector-quantized auto-encoders withadversarial training (VQ-GAN), which first extracts continuous frame-levelspeech representations with fine-grained codec, and extracts a discrete tokenfrom each continuous speech frame with coarse-grained codec; 2) we jointlyoptimize speech token, linguistic tokens, speaker token together with a largelanguage model and predict the discrete speech tokens autoregressively.Experiments show that FoundationTTS achieves a MOS gain of +0.14 compared tothe baseline system. In ASR customization tasks, our method achieves 7.09\% and10.35\% WERR respectively over two strong customized ASR baselines.

FoundationTTS: Text-to-Speech for ASR Customization with Generative Language Model