Edge0/Audio8-TTS-Realistic-Air modelining o'zbek tili uchun LoRA adapteri. Zero-shot voice cloning (reference audio + reference matn) saqlangan holda o'zbekcha gapirishga moslashtirilgan.
both AR qatlamlari, 2 epoxa, LR=0.0001.BaseLayer/uzbek-speech-40h-retranscript, Beehzod/uzbek_speech_data, BaseLayer/uzbek-target-speaker. Chetlaridagi jimlik kesilgan, 0.45 s dan uzun ichki pauzalar 0.30 s gacha qisqartirilgan. Reference — shu speakerning boshqa gapi (namunalarning ~90%).scale = 1.0. Sifat subyektiv tinglash bilan baholangan.adapters/step-N/ — oraliq checkpointlar; adapters/final/; resume/ — davom ettirish uchun to'liq checkpoint; logs/.import torch
from transformers import AutoModel, AutoProcessor
from peft import PeftModel
base = "Edge0/Audio8-TTS-Realistic-Air"
processor = AutoProcessor.from_pretrained(base, trust_remote_code=True)
model = AutoModel.from_pretrained(base, trust_remote_code=True, dtype=torch.bfloat16)
model = PeftModel.from_pretrained(model, "BaseLayer/audio8-uzbek-lora-v4")
model = model.merge_and_unload().eval().to("cuda")
inputs = processor(text=["Assalomu alaykum, bugun havo juda yaxshi."],
reference_audio=["ref.wav"], reference_text=["Reference'da aytilgan gap."], return_tensors="pt")
inputs = {k: v.to("cuda") for k, v in inputs.items()}
out = model.generate(**inputs, max_new_tokens=1024, temperature=0.8, top_p=0.95, top_k=50, do_sample=True, return_dict_in_generate=True)
wav, lens = model.decode_audio(out.codes)
Ovozni klonlashda ovoz egasining roziligi shart. O'qitish ma'lumotlarining litsenziyalarini tijoriy foydalanishdan oldin tekshiring.
resume/optimizer.pt 152 MBadapter_model.safetensors 76 MBadapters/final/adapter_model.safetensors 76 MBadapters/step-1000/adapter_model.safetensors 76 MBadapters/step-1500/adapter_model.safetensors 76 MBadapters/step-2000/adapter_model.safetensors 76 MBadapters/step-2500/adapter_model.safetensors 76 MBadapters/step-3000/adapter_model.safetensors 76 MBadapters/step-3080/adapter_model.safetensors 76 MBadapters/step-500/adapter_model.safetensors 76 MBresume/adapter_model.safetensors 76 MBresume/rng_state.pth 14.6 kBresume/training_args.bin 5.8 kBresume/scheduler.pt 1.5 kBresume/scaler.pt 1.4 kBEdge0/Audio8-TTS-Realistic-Air