File size: 3,723 Bytes
61383c3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
from __future__ import annotations

import json
import tempfile
import uuid
from pathlib import Path

import gradio as gr
import soundfile as sf

try:
    import spaces
except ImportError:
    class spaces:
        class GPU:
            def __init__(self, func=None, duration=60):
                self.func = func

            def __call__(self, *args, **kwargs):
                if self.func is not None:
                    return self.func(*args, **kwargs)
                return args[0]

from merit_runtime import compare_audio
from pyharp import ModelCard, build_endpoint


MIN_AUDIO_SECONDS = 5
MAX_AUDIO_SECONDS = 10
OUTPUT_ROOT = Path(tempfile.gettempdir()) / "merit_outputs"

model_card = ModelCard(
    name="MERIT",
    description=(
        "Compare two music clips using independent melody, rhythm, "
        "and timbre similarity representations."
    ),
    author="AMAAI Lab",
    tags=[
        "music-information-retrieval",
        "music-similarity",
        "melody",
        "rhythm",
        "timbre",
    ],
)


def _validate_audio(path: str | None, label: str) -> str:
    if not path:
        raise gr.Error(f"Please upload {label}.")

    try:
        duration = sf.info(path).duration
    except Exception as exc:
        raise gr.Error(f"Could not read {label}: {exc}") from exc

    if duration <= 0:
        raise gr.Error(f"{label} is empty.")
    if duration < MIN_AUDIO_SECONDS:
        raise gr.Error(
            f"{label} must be at least {MIN_AUDIO_SECONDS} seconds long. "
            f"Received {duration:.1f} seconds."
        )
    if duration > MAX_AUDIO_SECONDS:
        raise gr.Error(
            f"{label} must be no longer than {MAX_AUDIO_SECONDS} seconds. "
            f"Received {duration:.1f} seconds."
        )
    return path


@spaces.GPU(duration=120)
def process_fn(
    reference_audio: str | None,
    comparison_audio: str | None,
) -> str:
    reference_audio = _validate_audio(reference_audio, "Reference Audio")
    comparison_audio = _validate_audio(comparison_audio, "Comparison Audio")

    try:
        scores = compare_audio(reference_audio, comparison_audio)
    except Exception as exc:
        raise gr.Error(f"MERIT inference failed: {exc}") from exc

    output_dir = OUTPUT_ROOT / uuid.uuid4().hex
    output_dir.mkdir(parents=True, exist_ok=True)
    output_path = output_dir / "merit_similarity.json"
    output_path.write_text(
        json.dumps(
            {
                "model": "MERIT",
                "score_type": "cosine_similarity",
                "score_range": [-1.0, 1.0],
                "scores": scores,
            },
            indent=2,
        )
        + "\n",
        encoding="utf-8",
    )
    return str(output_path)


with gr.Blocks(title="MERIT Music Similarity") as demo:
    input_components = [
        gr.Audio(
            type="filepath",
            label="Reference Audio",
        )
        .harp_required(True)
        .set_info("First music clip, 5 to 10 seconds long."),
        gr.Audio(
            type="filepath",
            label="Comparison Audio",
        )
        .harp_required(True)
        .set_info("Second music clip, 5 to 10 seconds long."),
    ]
    output_components = [
        gr.File(
            type="filepath",
            file_types=[".json"],
            label="Similarity Results",
        ).set_info("Melody, rhythm, and timbre cosine similarity scores."),
    ]
    build_endpoint(
        model_card=model_card,
        input_components=input_components,
        output_components=output_components,
        process_fn=process_fn,
    )


if __name__ == "__main__":
    demo.queue(default_concurrency_limit=1).launch(
        show_error=True,
        pwa=True,
    )