SimpleTTS
Decoder-only transformer TTS, using WavTokenizer as the audio codec.
Samples
| Text | Audio |
|---|---|
| Sample 1 | |
| Sample 2 | |
| Sample 3 | |
| Sample 4 | |
| Sample 5 |
Setup
git clone https://github.com/FirstPotatoCoder/SimpleTTS.git
cd SimpleTTS
bash setup.sh
python scripts/download_weights.py
Run
python examples/run_inference.py "Hello, this is a quick test-run. I'm checking whether the model can handle longer input text without any issues. This sentence is roughly five times the length of the original short test phrase."
Or from Python:
from tts.inference import TTSPipeline
pipe = TTSPipeline(
tts_weights="weights/tts.pt",
wavtokenizer_weights="weights/wavtokenizer.ckpt",
wavtokenizer_config="configs/wavtokenizer_config.yaml",
)
pipe.generate("Some text to speak.", out_path="out.wav")
Limitations
- The model sometimes hallucinates, generating babbling or garbled output on rare or unseen words — likely due to the limited amount of training data.
- Works best when generating ~3s to 15s of audio, matching the length distribution of its training data.
- No chunking support yet — the current repo only supports clip-by-clip generation, one sample at a time.
Credits
Audio tokenizer/codec: WavTokenizer
(vendored under wavtokenizer/, trimmed to inference-only code).
Data synthesis: Kokoro, used to generate ~100 hours of synthetic audio for training this TTS model.
Text phonemization: espeak-ng, used to convert input text into phonemes.