ru-irina-1p57m/README.md
| 1 | # Root A raw fp16 package |
| 2 | |
| 3 | This package is the current duration + acoustic + decoder student stack as raw fp16 weights. |
| 4 | It is intended for a standalone runtime implementation, not PyTorch loading. |
| 5 | |
| 6 | - Parameters: 1,566,764 |
| 7 | - Weight blob: 3,133,528 bytes |
| 8 | - Asset bytes with phoneme config: 3,138,293 |
| 9 | - Duration length scale: 1 |
| 10 | - Runtime boundary: accepts Piper phoneme ID sequences, predicts durations, predicts Piper generator latents, then decodes waveform. |
| 11 | - Not complete arbitrary-text TTS yet: the Piper/eSpeak text frontend still has to be replaced or embedded. |
| 12 | |