Transformer-TTS and TransformerTTS

These two projects are competitors, as both are independent PyTorch implementations of the "Neural Speech Synthesis with Transformer Network" paper, aiming to provide a non-autoregressive Transformer-based neural network for text-to-speech.

Transformer-TTS
51
Established
TransformerTTS
51
Established
Maintenance 0/25
Adoption 10/25
Maturity 16/25
Community 25/25
Maintenance 0/25
Adoption 10/25
Maturity 16/25
Community 25/25
Stars: 690
Forks: 139
Downloads:
Commits (30d): 0
Language: Python
License: MIT
Stars: 1,161
Forks: 222
Downloads:
Commits (30d): 0
Language: Python
License:
Stale 6m No Package No Dependents
Archived Stale 6m No Package No Dependents

About Transformer-TTS

soobinseo/Transformer-TTS

A Pytorch Implementation of "Neural Speech Synthesis with Transformer Network"

Implements end-to-end mel-spectrogram synthesis using multi-head self-attention in both encoder and decoder, achieving 3-4x faster training than seq2seq baselines like Tacotron. Replaces the WaveNet vocoder with a CBHG-based postnet and Griffin-Lim vocoding for waveform reconstruction. Trained on LJSpeech with Noam-style learning rate scheduling and gradient clipping, with pretrained checkpoints available.

About TransformerTTS

spring-media/TransformerTTS

🤖💬 Transformer TTS: Implementation of a non-autoregressive Transformer based neural network for text to speech.

Built on TensorFlow 2, it combines a two-stage pipeline with an Aligner model for duration extraction and a Forward Transformer for parallel mel-spectrogram generation with controllable pitch prediction. Pre-trained LJSpeech weights integrate seamlessly with MelGAN and HiFiGAN vocoders for end-to-end synthesis, while the non-autoregressive approach eliminates repetition artifacts and enables real-time inference with speed/pitch modulation.

Scores updated daily from GitHub, PyPI, and npm data. How scores work