SpeechLM-based Text-to-Speech on LibriSpeech (960h)
10.5WERPURE Codec
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| PURE CodecCodec=PURE, enhancement_frontend=TF-GridNet2025.11 | 10.5 | 0.68 | 3.95 | |
| Baseline DACCodec=DAC2025.11 | 10.8 | 0.7 | 3.68 | |
| Abl.Benhancement_model_scaling=alternative frontend2025.11 | 11.1 | 0.71 | 3.99 | |
| Abl.Cenhancement_model_scaling=larger models2025.11 | 11.6 | 0.69 | 3.96 | |
| Abl.Aenhancement_model_scaling=smaller frontend2025.11 | 14 | 0.7 | 3.92 |