File size: 6,525 Bytes
4ca50b5
 
 
 
 
 
 
 
 
baab697
4ca50b5
 
baab697
4ca50b5
 
 
 
 
 
 
baab697
4ca50b5
 
 
 
 
 
 
 
 
 
 
 
 
 
baab697
4ca50b5
 
 
 
 
 
 
 
 
 
 
 
baab697
4ca50b5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
baab697
4ca50b5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
baab697
4ca50b5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
baab697
4ca50b5
 
 
 
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
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
# app.py
import gradio as gr
import librosa
import numpy as np
from openvino import runtime as ov
import soundfile as sf
import warnings
import os
from pathlib import Path

warnings.filterwarnings('ignore', category=FutureWarning)
warnings.filterwarnings('ignore', category=UserWarning)

def estimate_key(y, sr):
    """Estimate the musical key using chroma features."""
    chroma = librosa.feature.chroma_cqt(y=y, sr=sr)
    chroma_avg = np.mean(chroma, axis=1)
    keys = ['C', 'C#', 'D', 'D#', 'E', 'F', 'F#', 'G', 'G#', 'A', 'A#', 'B']
    key_index = np.argmax(chroma_avg)
    return keys[key_index]

def classify_instrument(spectral_centroid, rms_energy):
    """Classify instrument type based on spectral characteristics."""
    if spectral_centroid < 500:
        if rms_energy > 0.1:
            return "bass"
        return "sub"
    elif spectral_centroid < 2000:
        if rms_energy > 0.15:
            return "drums"
        return "perc"
    elif spectral_centroid < 4000:
        return "synth"
    else:
        return "high"

def get_musical_tempo_description(tempo):
    """Convert numerical tempo to musical description."""
    if tempo < 70:
        return "slow"
    elif tempo < 100:
        return "chill"
    elif tempo < 120:
        return "upbeat"
    elif tempo < 140:
        return "energetic"
    else:
        return "fast"

def generate_prompt(keys, avg_tempo, streams_info, genre="electronic"):
    """Generate a concise, Suno-friendly prompt under 200 characters."""
    most_common_key = max(set(keys), key=keys.count) if keys else "C"
    
    instrument_counts = {}
    for info in streams_info:
        inst_type = info['type']
        instrument_counts[inst_type] = instrument_counts.get(inst_type, 0) + 1
    
    main_elements = [k for k, v in sorted(instrument_counts.items(), key=lambda x: x[1], reverse=True)[:2]]
    tempo_desc = get_musical_tempo_description(avg_tempo)
    
    prompt = f"{most_common_key} {int(avg_tempo)}bpm {tempo_desc} {genre} with {' + '.join(main_elements)}, dark atmosphere + reverb"
    
    if len(prompt) > 200:
        prompt = prompt[:197] + "..."
        
    return prompt

def process_audio(audio_path, genre):
    """Process audio file and generate prompt."""
    try:
        # Load audio
        y, sr = librosa.load(audio_path, sr=None)
        print(f"Audio loaded: {len(y)} samples, Sample rate: {sr}")
        
        # Configure OpenVINO model
        model_path = os.path.join(os.path.dirname(__file__), "models", "htdemucs_v4.xml")
        core = ov.Core()
        model = core.read_model(model_path)
        compiled_model = core.compile_model(model, "CPU")
        
        input_node = compiled_model.input(0)
        output_node = compiled_model.output(0)
        target_shape = (1, 4, 2048, 336)
        
        total_size = np.prod(target_shape)
        if len(y) < total_size:
            input_data = np.pad(y, (0, total_size - len(y)), mode='constant')
        else:
            input_data = y[:total_size]
        
        input_data = input_data.reshape(target_shape).astype(np.float32)
        input_tensor = ov.Tensor(input_data)
        
        outputs = compiled_model([input_tensor])[output_node]
        separated_audios = outputs[0]
        
        # Analysis lists
        keys = []
        avg_tempos = []
        streams_info = []
        
        # Create temporary directory for separated streams
        temp_dir = Path("temp_streams")
        temp_dir.mkdir(exist_ok=True)
        
        # Process each separated audio stream
        for i in range(separated_audios.shape[0]):
            stream = separated_audios[i].reshape(-1)
            
            try:
                output_file = temp_dir / f'separated_stream_{i+1}.wav'
                sf.write(str(output_file), stream, sr)
                
                y_s, sr_s = librosa.load(str(output_file), sr=None)
                
                if len(y_s) < sr_s * 0.1:
                    continue
                
                # Calculate audio features
                tempo_s, _ = librosa.beat.beat_track(y=y_s, sr=sr_s)
                spectral_centroid_s = np.mean(librosa.feature.spectral_centroid(y=y_s, sr=sr_s))
                rms_s = np.mean(librosa.feature.rms(y=y_s))
                key_s = estimate_key(y_s, sr_s)
                
                # Store all information
                streams_info.append({
                    'type': classify_instrument(spectral_centroid_s, rms_s),
                    'centroid': spectral_centroid_s,
                    'energy': rms_s
                })
                
                keys.append(key_s)
                avg_tempos.append(tempo_s)
                
            except Exception as e:
                print(f"Warning: Could not process stream {i+1}: {str(e)}")
                continue
            finally:
                # Clean up temporary file
                if output_file.exists():
                    output_file.unlink()
        
        # Clean up temporary directory
        temp_dir.rmdir()
        
        if len(avg_tempos) > 0:
            avg_tempo = np.mean(avg_tempos)
            prompt = generate_prompt(keys, avg_tempo, streams_info, genre)
            return prompt, f"Character count: {len(prompt)}"
        else:
            return "Error: No valid audio streams were processed.", "Processing failed"
            
    except Exception as e:
        return f"Error processing the file: {str(e)}", "Processing failed"

# Create Gradio interface
def create_interface():
    genre_choices = ["electronic", "ambient", "trap", "synthwave", "house", "techno"]
    
    iface = gr.Interface(
        fn=process_audio,
        inputs=[
            gr.Audio(type="filepath", label="Upload Audio File"),
            gr.Dropdown(choices=genre_choices, label="Select Genre", value="electronic")
        ],
        outputs=[
            gr.Textbox(label="Generated Prompt"),
            gr.Textbox(label="Status")
        ],
        title="Audio Analysis to Suno Prompt Generator",
        description="Upload an audio file to generate a Suno-compatible prompt based on its musical characteristics.",
        examples=[],
        cache_examples=False
    )
    return iface

# Launch the interface
if __name__ == "__main__":
    iface = create_interface()
    iface.launch()