FastThaiG2P Achieves 0.15ms Latency for Thai Speech Pipelines

FastThaiG2P enables sub-millisecond Thai grapheme-to-phoneme conversion with an average latency of 0.15ms per utterance. The system utilizes a PyThaiNLP-tokenized dictionary and achieves a 0.5% OOV rate, supporting 82M-parameter StyleTTS 2 models via ONNX CPU inference.