Flare-TTS-28M

🎙️ Flare-TTS 28M

Welcome to Flare-TTS 28M, an open-source text-to-speech model with 28 million parameters trained on LJSpeech.

Quality and results

This model is okayish quality but it still sounds a bit robotish but you can clearly understand what the model tries to say. See this model as a proof-of-concept or a first-beta. Example: <audio controls src="https://cdn-uploads.huggingface.co/production/uploads/697f2832c2c5e4daa93cece7/vluuHSnp9Ietk7Uk1-hvG.mpga"></audio>

Training process

We trained this model for ~300 epochs on a single A6000 GPU for ~24 hours. The full training code can be found in this repo as start.sh and train.py. Just run start.sh to train this model yourself.

Architecture

This model was trained using CoquiTTS. For the architecture we chose GlowTTS.

Training dataset

We trained on the full LJSpeech dataset. Thanks to keithito for this :-)

How to use

As soon as you have the model checkpoint (model.pth) and config.json on your device, you can generate a sample using:

tts --text "Hello world, this is my first trained TTS model." \
    --model_path model.pth \
    --config_path config.json \
    --out_path output_1.wav

Final thoughts

We don't think it's perfect - it's more like a proof of concept. So please do not use this model for production use cases but more for experiments. We are happy to share more of this soon - stay tuned for Flare-TTS v2 :D