Autoregressive Speech Synthesis without Vector Quantization
This addresses the robustness and fidelity issues in speech synthesis for applications requiring high-quality audio, though it is an incremental improvement over existing autoregressive approaches.
The paper tackles the problem of text-to-speech synthesis by introducing MELLE, a method that autoregressively generates continuous mel-spectrogram frames directly from text, bypassing vector quantization, and achieves superior performance compared to existing two-stage models like VALL-E across multiple metrics.
We present MELLE, a novel continuous-valued token based language modeling approach for text-to-speech synthesis (TTS). MELLE autoregressively generates continuous mel-spectrogram frames directly from text condition, bypassing the need for vector quantization, which is typically designed for audio compression and sacrifices fidelity compared to continuous representations. Specifically, (i) instead of cross-entropy loss, we apply regression loss with a proposed spectrogram flux loss function to model the probability distribution of the continuous-valued tokens; (ii) we have incorporated variational inference into MELLE to facilitate sampling mechanisms, thereby enhancing the output diversity and model robustness. Experiments demonstrate that, compared to the two-stage codec language model VALL-E and its variants, the single-stage MELLE mitigates robustness issues by avoiding the inherent flaws of sampling vector-quantized codes, achieves superior performance across multiple metrics, and, most importantly, offers a more streamlined paradigm. The demos of our work are provided at https://aka.ms/melle.