# TalkNet: Fully-Convolutional Non-Autoregressive Speech Synthesis Model

arXiv: https://arxiv.org/abs/2005.05514 · HTML: https://mlsanity.com/paper/2005.05514/

## Key facts

- Cited by 9 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 12 May 2020.

## Built on by (most cited first, 9 of 9)

- [A Survey on Neural Speech Synthesis](https://mlsanity.com/paper/2106.15561/) (2021)
- [INFP: Audio-Driven Interactive Head Generation in Dyadic Conversations](https://mlsanity.com/paper/2412.04037/) (2024)
- [Review of end-to-end speech synthesis technology based on deep learning](https://mlsanity.com/paper/2104.09995/) (2021)
- [Transformer-Based Speech Synthesizer Attribution in an Open Set Scenario](https://mlsanity.com/paper/2210.07546/) (2022)
- [Multi-human Interactive Talking Dataset](https://mlsanity.com/paper/2508.03050/) (2025)
- [InterDyad: Interactive Dyadic Speech-to-Video Generation by Querying Intermediate Visual Guidance](https://mlsanity.com/paper/2603.23132/) (2026)
- [Sprachsynthese -- State-of-the-Art in englischer und deutscher Sprache](https://mlsanity.com/paper/2106.06230/) (2021)
- [Fitting Different Interactive Information: Joint Classification of Emotion and Intention](https://mlsanity.com/paper/2501.06215/) (2025)
- [LRW-Persian: Lip-reading in the Wild Dataset for Persian Language](https://mlsanity.com/paper/2510.22716/) (2025)

## Abstract (arXiv)

We propose TalkNet, a convolutional non-autoregressive neural model for speech synthesis. The model consists of two 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 generates a mel-spectrogram from the expanded text. To train a grapheme duration predictor, we add the grapheme duration to the training dataset using a pre-trained Connectionist Temporal Classification (CTC)-based speech recognition model. The explicit duration prediction eliminates word skipping and repeating. Experiments on the LJSpeech dataset show that the speech quality nearly matches auto-regressive models. The model is very compact -- it has 10.8M parameters, almost 3x less than the present state-of-the-art text-to-speech models. The non-autoregressive architecture allows for fast training and inference.