Implementation of E2 TTS: Embarrassingly Easy Fully Non-Autoregressive Zero-Shot TTS

Usage:

build filelist and phoneset with scripts/build_ljspeech.py and build_filelist.py

update the filelist path and phoneset path config/e2_tts.yaml

train model: python3 train.py -c config/e2_tts.yaml

inference with rfwave vocoder: download acoustic ckpt trained with LJSpeech and LibriTTS, the vocoder ckpt trained with LibriTTS and inference with inference.py. Synthesized samples from this checkpoint can be found here

python3 inference.py --test_txt tests/test.txt \
    --aco_model_dir /path/to/e2_tts-cfg_2_cond2-bf16-large_batch \
    --voc_model_dir /path/to/rfwave-libritts-24k \
    --phoneset tests/phoneset.th \
    --save_dir syn_e2-08-22 --sr 24000

Some discussion can be found here.