File size: 4,419 Bytes
14e7761
 
 
 
 
 
 
 
 
 
 
153d7d2
14e7761
 
 
153d7d2
 
 
 
 
 
 
 
 
 
 
14e7761
 
153d7d2
14e7761
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6e103ce
 
 
 
 
 
 
 
 
 
 
 
14e7761
 
 
6e103ce
 
 
 
 
 
 
 
 
 
 
 
14e7761
 
 
6e103ce
 
 
 
 
 
 
 
 
 
 
 
14e7761
 
 
 
 
 
 
153d7d2
14e7761
6e103ce
 
14e7761
 
6e103ce
 
14e7761
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
153d7d2
 
14e7761
 
153d7d2
 
14e7761
 
 
 
153d7d2
14e7761
 
 
 
153d7d2
14e7761
 
 
153d7d2
14e7761
 
 
 
 
 
 
 
 
 
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
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
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 model import EvalNet
from utils import get_modelist, find_wav_files, embed_img


TRANSLATE = {
    "vibrato": "Rou xian",
    "trill": "Chan yin",
    "tremolo": "Chan gong",
    "staccato": "Dun gong",
    "ricochet": "Pao gong",
    "pizzicato": "Bo xian",
    "percussive": "Ji gong",
    "legato_slide_glissando": "Lian hua yin",
    "harmonic": "Fan yin",
    "diangong": "Dian gong",
    "detache": "Fen gong",
}
CLASSES = list(TRANSLATE.keys())
TEMP_DIR = "./__pycache__/tmp"
SAMPLE_RATE = 44100


def circular_padding(y: np.ndarray, sr: int, dur=3):
    if len(y) >= sr * dur:
        return y[: sr * dur]

    size = sr * dur // len(y) + int((sr * dur) % len(y) > 0)
    arrays = []
    for _ in range(size):
        arrays.append(y)

    y = np.hstack(arrays)
    return y[: sr * dur]


def wav2mel(audio_path: str):
    y, sr = librosa.load(audio_path, sr=SAMPLE_RATE)
    y = circular_padding(y, 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()


def wav2cqt(audio_path: str):
    y, sr = librosa.load(audio_path, sr=SAMPLE_RATE)
    y = circular_padding(y, 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()


def wav2chroma(audio_path: str):
    y, sr = librosa.load(audio_path, sr=SAMPLE_RATE)
    y = circular_padding(y, 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()


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, "Please input an audio!"

    spec = log_name.split("_")[-3]
    os.makedirs(folder_path, exist_ok=True)
    try:
        model = EvalNet(log_name, len(TRANSLATE)).model
        eval("wav2%s" % spec)(wav_path)

    except Exception as e:
        return None, f"{e}"

    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_wavs = find_wav_files()
    model_num = len(models)
    for wav in example_wavs:
        examples.append([wav, models[random.randint(0, model_num - 1)]])

    with gr.Blocks() as demo:
        gr.Interface(
            fn=infer,
            inputs=[
                gr.Audio(label="Upload a recording", type="filepath"),
                gr.Dropdown(choices=models, label="Select a model", value=models[0]),
            ],
            outputs=[
                gr.Textbox(label="Audio filename", show_copy_button=True),
                gr.Textbox(label="Playing tech recognition", show_copy_button=True),
            ],
            examples=examples,
            cache_examples=False,
            allow_flagging="never",
            title="It is recommended to keep the recording length around 3s.",
        )

        gr.Markdown(
            """
# 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()