High Quality Streaming Speech Synthesis with Low, Sentence-Length-Independent Latency

TL;DR

This paper introduces a low-latency, end-to-end speech synthesis system capable of real-time performance on CPUs, using a novel attention mechanism and streaming inference to produce near-natural speech quickly regardless of sentence length.

Contribution

The paper proposes a new autoregressive TTS architecture with location-based attention and streaming inference, achieving low latency and high-quality speech synthesis.

Findings

Achieves 31x real-time speed on CPU

Maintains nearly constant latency regardless of sentence length

Produces speech of almost natural quality

Abstract

This paper presents an end-to-end text-to-speech system with low latency on a CPU, suitable for real-time applications. The system is composed of an autoregressive attention-based sequence-to-sequence acoustic model and the LPCNet vocoder for waveform generation. An acoustic model architecture that adopts modules from both the Tacotron 1 and 2 models is proposed, while stability is ensured by using a recently proposed purely location-based attention mechanism, suitable for arbitrary sentence length generation. During inference, the decoder is unrolled and acoustic feature generation is performed in a streaming manner, allowing for a nearly constant latency which is independent from the sentence length. Experimental results show that the acoustic model can produce feature sequences with minimal latency about 31 times faster than real-time on a computer CPU and 6.5 times on a mobile CPU, enabling it to meet the conditions required for real-time applications on both devices. The full end-to-end system can generate almost natural quality speech, which is verified by listening tests.