TalkNet 2: Non-Autoregressive Depth-Wise Separable Convolutional Model for Speech Synthesis with Explicit Pitch and Duration Prediction

TL;DR

TalkNet 2 introduces a non-autoregressive, efficient convolutional model for speech synthesis that explicitly predicts pitch and duration, achieving high-quality speech with fewer parameters and faster inference.

Contribution

The paper presents TalkNet 2, a novel non-autoregressive convolutional model with explicit pitch and duration prediction for improved speech synthesis efficiency and quality.

Findings

Achieves MOS 4.08 on LJSpeech, close to state-of-the-art.

Uses only 13.2M parameters, nearly half of comparable models.

Enables fast training and inference suitable for embedded devices.

Abstract

We propose TalkNet, a non-autoregressive convolutional neural model for speech synthesis with explicit pitch and duration prediction. The model consists of three feed-forward convolutional networks. The first network predicts grapheme durations. An input text is expanded by repeating each symbol according to the predicted duration. The second network predicts pitch value for every mel frame. The third network generates a mel-spectrogram from the expanded text conditioned on predicted pitch. All networks are based on 1D depth-wise separable convolutional architecture. The explicit duration prediction eliminates word skipping and repeating. The quality of the generated speech nearly matches the best auto-regressive models - TalkNet trained on the LJSpeech dataset got MOS 4.08. The model has only 13.2M parameters, almost 2x less than the present state-of-the-art text-to-speech models. The non-autoregressive architecture allows for fast training and inference. The small model size and fast inference make the TalkNet an attractive candidate for embedded speech synthesis.