Chinese Mandarin text-to-speech
A modular PyTorch framework for training and running Chinese Mandarin multi-speaker TTS, built around a FastSpeech2 acoustic model with configurable neural vocoders.
Same problem, same approach. Swapping one for another is a config change, not a rewrite.
Choose it for an actively maintained, pip-installable multi-voice engine with Chinese support and prompt/emotion control.
EmotiVoice 😊: a Multi-Voice and Prompt-Controlled TTS Engine
Solves the same problem with a different architecture or at a different layer. Expect to rewrite the integration.
Pick it if your stack is TensorFlow 2 and you want ready Chinese checkpoints, at the cost of an inactive repo.
:stuck_out_tongue_closed_eyes: TensorFlowTTS: Real-Time State-of-the-art Speech Synthesis for Tensorflow 2 (supported including English, French, Korean, Chinese, German and Easy to adapt for other languages)
Choose it for end-to-end Chinese VITS with ONNX streaming instead of a FastSpeech2+vocoder pipeline.
Best practice TTS based on BERT and VITS with some Natural Speech Features Of Microsoft; Support ONNX streaming out!
Choose it when you need few-shot voice cloning from seconds of audio rather than a fixed Chinese multi-speaker model.
🚀Clone a voice in 5 seconds to generate arbitrary speech in real-time
Choose it for end-to-end VITS quality with a single-stage pipeline, accepting English-centric data and no maintenance.
VITS: Conditional Variational Autoencoder with Adversarial Learning for End-to-End Text-to-Speech
Same job on PaddlePaddle with many model recipes, but archived and with no aishell3-focused Chinese recipe.
PAddle PARAllel text-to-speech toolKIT (supporting Tacotron2, Transformer TTS, FastSpeech2/FastPitch, SpeedySpeech, WaveFlow and Parallel WaveGAN)
Choose it for zero-shot cloning from a short sample, at the cost of large weights and an unmaintained repo.
Foundational model for human-like, expressive TTS
Choose it for a permissively licensed Whisper-inverted pipeline, though it is not tuned for Chinese multi-speaker data.
An Open Source text-to-speech system built by inverting Whisper.
Choose it when you want few-shot English voice cloning with a ready toolbox UI instead of a training framework.
Clone a voice in 5 seconds to generate arbitrary speech in real-time
Choose it for high-quality multi-voice English synthesis, accepting slow inference and no Chinese support.
A multi-voice TTS system trained with an emphasis on quality
An earlier or more limited way to do the job. Still the right call when you need something small, proven, or CPU-only.
Clean minimal FastSpeech2 reference, but unmaintained for ~3 years and without a multi-speaker Chinese training recipe.
An implementation of Microsoft's "FastSpeech 2: Fast and High-Quality End-to-End Text to Speech"
Autoregressive WaveRNN TTS+vocoder, slower and English-only; a swap only for a small readable legacy pipeline.
WaveRNN Vocoder + TTS
Tacotron plus global style tokens: older seq2seq plus separate vocoder, useful only for style-transfer experiments.
A PyTorch implementation of Style Tokens: Unsupervised Style Modeling, Control and Transfer in End-to-End Speech Synthesis
Pre-FastSpeech convolutional seq2seq TTS, unmaintained; only for reproducing an old baseline.
PyTorch implementation of convolutional neural networks-based text-to-speech synthesis models
First-generation FastSpeech: cheaper to train but lower quality and long unmaintained.
The Implementation of FastSpeech based on pytorch.
Inference-only ONNX engine for fast local/CPU deployment; no training code and now archived.
A fast, local neural text to speech system
TensorFlow 1 Tacotron with a pre-trained English model; only for reproducing a legacy baseline.
A TensorFlow implementation of Google's Tacotron speech synthesis with pre-trained model (unofficial)
Russian/English Tacotron voice cloning, archived and narrow; only if you need that language pair.
Phoneme multilingual(Russian-English) voice cloning based on
These projects were analysed and named mandarin-tts among their alternatives. The relationship is not symmetric — how mandarin-tts rates them is a separate judgement, made when mandarin-tts is analysed in its own right.
calls mandarin-tts “choose for a Chinese/Mandarin FastSpeech2 implementation specifically, not a general toolkit.”
:robot: :speech_balloon: Deep learning for Text to Speech (Discussion forum: https://discourse.mozilla.org/c/tts)
calls mandarin-tts “Choose for a Mandarin-specific FastSpeech2/WaveGlow TTS pipeline, but it's an older, unmaintained repo.”
Desktop application for neural speech synthesis written in C++