please, give me some suggestions To explain, I type "Sleep Music For Deep Sleep" and it will return a wav file as https://www.youtube.com/watch?v=1wAdQhFJy54 or I give a wav file, it will return the same Now here is what I have tried: https://github.com/facebookresearch/audiocraft but it has a Low Quality import torchaudio from audiocraft.models import MusicGen au_crmode = MusicGen.get_pretrained(pre, cache_dir=pret_loca) au_crmode.set_generation_params(duration=kwargs['thoigian']) melody, sr = torchaudio.load('./176_183.wav') # generates using the melody from the given audio and the provided descriptions. wav = au_crmode.generate_with_chroma(input_text, melody[None].expand(3, -1, -1), sr) for idx, one_wav in enumerate(wav): audio_write(f'{idx}', one_wav.cpu(), au_crmode.sample_rate, strategy="loudness") or processor = AutoProcessor.from_pretrained(pretra[0], cache_dir=pret_loca) model = MusicgenForConditionalGeneration.from_pretrained(pretra[0], cache_dir=pret_loca) sample_length = 30 num_tokens = sample_length * model.config.audio_encoder.frame_rate + 3 inputs = processor( text=input_text, padding=True, return_tensors="pt", ).to(device) audio_values = model.generate( **inputs, max_new_tokens=num_tokens, # defines the length of the generated music piece guidance_scale=g_scale, # controls the creativity level # The guidance_scale is used in classifier free guidance...Higher guidance scale encourages the model to generate samples that are more closely linked to the input prompt do_sample=True, # enables stochastic sampling, making the generation process more creative # sampling leads to significantly better results than greedy (do_sample=False) ) https://github.com/yy1lab/Lyrics-Conditioned-Neural-Melody-Generation I can't run whether in python 2.7 or newer https://github.com/lucidrains/musiclm-pytorch it seems models in hubert have errors; in my case, they are KeyError: 'hubert_pretraining' with HuBERT Base or 'NoneType' object has no attribute 'task' with HuBERT Large any ideals Continue reading...
