File size: 2,555 Bytes
f67b703 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 |
import gradio as gr
import torchaudio
import torch
import os
from rave import RAVE # Assuming rave.py or pip package is available
from huggingface_hub import hf_hub_download
# ✅ Available RAVE models (can expand dynamically from HF repo)
RAVE_MODELS = {
"Guitar": "guitar_iil_b2048_r48000_z16.ts",
"Soprano Sax": "sax_soprano_franziskaschroeder_b2048_r48000_z20.ts",
"Organ (Archive)": "organ_archive_b2048_r48000_z16.ts",
"Organ (Bach)": "organ_bach_b2048_r48000_z16.ts",
"Voice Multivoice": "voice-multi-b2048-r48000-z11.ts",
"Birds Dawn Chorus": "birds_dawnchorus_b2048_r48000_z8.ts",
"Magnets": "magnets_b2048_r48000_z8.ts",
"Whale Songs": "humpbacks_pondbrain_b2048_r48000_z20.ts"
}
MODEL_CACHE = {}
def load_rave_model(model_name):
"""Load a RAVE model from Hugging Face or cache."""
if model_name in MODEL_CACHE:
return MODEL_CACHE[model_name]
model_file = hf_hub_download(
repo_id="Intelligent-Instruments-Lab/rave-models",
filename=RAVE_MODELS[model_name]
)
model = RAVE.load(model_file) # RAVE.load assumes wrapper for loading .ts file
model.eval()
MODEL_CACHE[model_name] = model
return model
def apply_rave(audio, model_name):
"""Apply selected RAVE style transfer model to uploaded audio."""
model = load_rave_model(model_name)
# Convert numpy audio (from Gradio) to torch tensor
audio_tensor = torch.tensor(audio[0]).unsqueeze(0) # [1, samples]
sr = audio[1]
if sr != 48000:
audio_tensor = torchaudio.functional.resample(audio_tensor, sr, 48000)
sr = 48000
# Pass through model (encode -> decode)
with torch.no_grad():
z = model.encode(audio_tensor)
processed_audio = model.decode(z)
processed_audio = processed_audio.squeeze().cpu().numpy()
return (processed_audio, sr)
# 🎛 Gradio Interface
with gr.Blocks() as demo:
gr.Markdown("## 🎛 RAVE Style Transfer on Stems")
gr.Markdown("Upload audio, select a RAVE model, and get a transformed version.")
with gr.Row():
audio_input = gr.Audio(type="numpy", label="Upload Audio", sources=["upload", "microphone"])
model_selector = gr.Dropdown(list(RAVE_MODELS.keys()), label="Select Style", value="Guitar")
with gr.Row():
output_audio = gr.Audio(type="numpy", label="Transformed Audio")
# API + UI trigger
process_btn = gr.Button("Apply Style Transfer")
process_btn.click(fn=apply_rave, inputs=[audio_input, model_selector], outputs=output_audio)
demo.launch()
|