Spaces:
Running
Running
import os | |
import torch | |
import random | |
import shutil | |
import librosa | |
import warnings | |
import numpy as np | |
import gradio as gr | |
import librosa.display | |
import matplotlib.pyplot as plt | |
from utils import get_modelist, find_audio_files, embed_img, TEMP_DIR | |
from model import EvalNet | |
TRANSLATE = { | |
"Gong": "宫", | |
"Shang": "商", | |
"Jue": "角", | |
"Zhi": "徵", | |
"Yu": "羽", | |
} | |
CLASSES = list(TRANSLATE.keys()) | |
SAMPLE_RATE = 44100 | |
def zero_padding(y: np.ndarray, end: int): | |
size = len(y) | |
if size < end: | |
return np.concatenate((y, np.zeros(end - size))) | |
elif size > end: | |
return y[-end:] | |
return y | |
def audio2mel(audio_path: str, seg_len=20): | |
os.makedirs(TEMP_DIR, exist_ok=True) | |
try: | |
y, sr = librosa.load(audio_path, sr=SAMPLE_RATE) | |
y = zero_padding(y, seg_len * sr) | |
mel_spec = librosa.feature.melspectrogram(y=y, sr=sr) | |
log_mel_spec = librosa.power_to_db(mel_spec, ref=np.max) | |
librosa.display.specshow(log_mel_spec) | |
plt.axis("off") | |
plt.savefig( | |
f"{TEMP_DIR}/output.jpg", | |
bbox_inches="tight", | |
pad_inches=0.0, | |
) | |
plt.close() | |
except Exception as e: | |
print(f"Error converting {audio_path} : {e}") | |
def audio2cqt(audio_path: str, seg_len=20): | |
os.makedirs(TEMP_DIR, exist_ok=True) | |
try: | |
y, sr = librosa.load(audio_path, sr=SAMPLE_RATE) | |
y = zero_padding(y, seg_len * sr) | |
cqt_spec = librosa.cqt(y=y, sr=sr) | |
log_cqt_spec = librosa.power_to_db(np.abs(cqt_spec) ** 2, ref=np.max) | |
librosa.display.specshow(log_cqt_spec) | |
plt.axis("off") | |
plt.savefig( | |
f"{TEMP_DIR}/output.jpg", | |
bbox_inches="tight", | |
pad_inches=0.0, | |
) | |
plt.close() | |
except Exception as e: | |
print(f"Error converting {audio_path} : {e}") | |
def audio2chroma(audio_path: str, seg_len=20): | |
os.makedirs(TEMP_DIR, exist_ok=True) | |
try: | |
y, sr = librosa.load(audio_path, sr=SAMPLE_RATE) | |
y = zero_padding(y, seg_len * sr) | |
chroma_spec = librosa.feature.chroma_stft(y=y, sr=sr) | |
log_chroma_spec = librosa.power_to_db(np.abs(chroma_spec) ** 2, ref=np.max) | |
librosa.display.specshow(log_chroma_spec) | |
plt.axis("off") | |
plt.savefig( | |
f"{TEMP_DIR}/output.jpg", | |
bbox_inches="tight", | |
pad_inches=0.0, | |
) | |
plt.close() | |
except Exception as e: | |
print(f"Error converting {audio_path} : {e}") | |
def infer(wav_path: str, log_name: str, folder_path=TEMP_DIR): | |
if os.path.exists(folder_path): | |
shutil.rmtree(folder_path) | |
if not wav_path: | |
return None, "请输入音频 Please input an audio!" | |
try: | |
model = EvalNet(log_name, len(TRANSLATE)).model | |
except Exception as e: | |
return None, f"{e}" | |
spec = log_name.split("_")[-3] | |
eval("audio2%s" % spec)(wav_path) | |
input = embed_img(f"{folder_path}/output.jpg") | |
output: torch.Tensor = model(input) | |
pred_id = torch.max(output.data, 1)[1] | |
return ( | |
os.path.basename(wav_path), | |
f"{TRANSLATE[CLASSES[pred_id]]} ({CLASSES[pred_id].capitalize()})", | |
) | |
if __name__ == "__main__": | |
warnings.filterwarnings("ignore") | |
models = get_modelist() | |
examples = [] | |
example_audios = find_audio_files() | |
model_num = len(models) | |
for audio in example_audios: | |
examples.append([audio, models[random.randint(0, model_num - 1)]]) | |
with gr.Blocks() as demo: | |
gr.Interface( | |
fn=infer, | |
inputs=[ | |
gr.Audio(label="上传录音 Upload a recording", type="filepath"), | |
gr.Dropdown( | |
choices=models, label="选择模型 Select a model", value=models[0] | |
), | |
], | |
outputs=[ | |
gr.Textbox(label="音频文件名 Audio filename", show_copy_button=True), | |
gr.Textbox( | |
label="中国五声调式识别 Chinese pentatonic mode recognition", | |
show_copy_button=True, | |
), | |
], | |
examples=examples, | |
cache_examples=False, | |
flagging_mode="never", | |
title="建议录音时长保持在 20s 左右<br>It is recommended to keep the recording length around 20s.", | |
) | |
gr.Markdown( | |
""" | |
# 引用 Cite | |
```bibtex | |
@dataset{zhaorui_liu_2021_5676893, | |
author = {Monan Zhou, Shenyang Xu, Zhaorui Liu, Zhaowen Wang, Feng Yu, Wei Li and Baoqiang Han}, | |
title = {CCMusic: an Open and Diverse Database for Chinese Music Information Retrieval Research}, | |
month = {mar}, | |
year = {2024}, | |
publisher = {HuggingFace}, | |
version = {1.2}, | |
url = {https://huggingface.co/ccmusic-database} | |
} | |
```""" | |
) | |
demo.launch() | |