File size: 6,918 Bytes
d57e374
117486a
d57e374
117486a
 
4cf73d6
075c9a6
4cf73d6
 
117486a
075c9a6
117486a
 
 
1834911
 
d57e374
117486a
 
d57e374
117486a
 
d57e374
 
075c9a6
117486a
 
 
 
 
 
 
d57e374
075c9a6
d57e374
117486a
d57e374
 
 
 
 
 
 
 
 
 
 
075c9a6
117486a
d57e374
 
 
 
 
117486a
d57e374
 
 
 
 
 
075c9a6
55f08a9
 
 
 
117486a
 
 
 
 
d57e374
 
117486a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
55f08a9
075c9a6
117486a
 
075c9a6
d57e374
075c9a6
117486a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d57e374
075c9a6
117486a
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
182
183
184
185
186
187
188
import os
import gc
import torch
import shutil
import atexit
import torchaudio
import numpy as np
import gradio as gr
from pipeline.morph_pipeline_successed_ver1 import AudioLDM2MorphPipeline 
os.environ["CUDA_VISIBLE_DEVICES"] = "6"
# Initialize AudioLDM2 Pipeline
torch.cuda.set_device(0)
dtype = torch.float32
pipeline = AudioLDM2MorphPipeline.from_pretrained("cvssp/audioldm2-large", torch_dtype=dtype)
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
pipeline.to(device)


def morph_audio(audio_file1, audio_file2, num_inference_steps, prompt1='', prompt2='', negative_prompt1="Low quality", negative_prompt2="Low quality"):
    save_lora_dir = "output"
    if os.path.exists(save_lora_dir):
        shutil.rmtree(save_lora_dir)
    os.makedirs(save_lora_dir, exist_ok=True)
    
    # Load audio and compute duration
    waveform1, sample_rate1 = torchaudio.load(audio_file1)
    duration1 = waveform1.shape[1] / sample_rate1
    waveform2, sample_rate2 = torchaudio.load(audio_file2)
    duration2 = waveform2.shape[1] / sample_rate2

    # Compare durations and take the shorter one
    duration = int(min(duration1, duration2))
    
    # Perform morphing using the pipeline
    _ = pipeline(
        dtype = dtype,
        audio_file=audio_file1,
        audio_file2=audio_file2,
        audio_length_in_s=duration,
        time_pooling=2,
        freq_pooling=2,
        prompt_1=prompt1,
        prompt_2=prompt2,
        negative_prompt_1=negative_prompt1,
        negative_prompt_2=negative_prompt2,
        save_lora_dir=save_lora_dir,
        use_adain=True,
        use_reschedule=False,
        num_inference_steps=num_inference_steps,
        lamd=0.6,
        output_path=save_lora_dir,
        num_frames=5,
        fix_lora=None,
        use_lora=True,
        lora_steps=2,
        noisy_latent_with_lora=True,
        morphing_with_lora=True,
        use_morph_prompt=True,
        guidance_scale=7.5,
    )
    
    # Collect the output file paths
    output_paths = sorted(
    [os.path.join(save_lora_dir, file) for file in os.listdir(save_lora_dir) if file.endswith(".wav")],
    key=lambda x: int(os.path.splitext(os.path.basename(x))[0])
    )
    del waveform1, waveform2, _
    torch.cuda.empty_cache()
    gc.collect()


    return output_paths

def morph_audio_with_morphing_factor(audio_file1, audio_file2, alpha,num_inference_steps, prompt1='', prompt2='', negative_prompt1="Low quality", negative_prompt2="Low quality"):
    save_lora_dir = "output"
    if os.path.exists(save_lora_dir):
        shutil.rmtree(save_lora_dir)
    os.makedirs(save_lora_dir, exist_ok=True)
    
    # Load audio and compute duration
    waveform1, sample_rate1 = torchaudio.load(audio_file1)
    duration1 = waveform1.shape[1] / sample_rate1
    waveform2, sample_rate2 = torchaudio.load(audio_file2)
    duration2 = waveform2.shape[1] / sample_rate2

    # Compare durations and take the shorter one
    duration = int(min(duration1, duration2))
    try:
        # Perform morphing using the pipeline
        _ = pipeline(
            dtype = dtype,
            morphing_factor = alpha,
            audio_file=audio_file1,
            audio_file2=audio_file2,
            audio_length_in_s=duration,
            time_pooling=2,
            freq_pooling=2,
            prompt_1=prompt1,
            prompt_2=prompt2,
            negative_prompt_1=negative_prompt1,
            negative_prompt_2=negative_prompt2,
            save_lora_dir=save_lora_dir,
            use_adain=True,
            use_reschedule=False,
            num_inference_steps=num_inference_steps,
            lamd=0.6,
            output_path=save_lora_dir,
            num_frames=5,
            fix_lora=None,
            use_lora=True,
            lora_steps=2,
            noisy_latent_with_lora=True,
            morphing_with_lora=True,
            use_morph_prompt=True,
            guidance_scale=7.5,
        )
        output_paths = os.path.join(save_lora_dir, 'interpolated.wav')
    
    except RuntimeError as e:
        if "CUDA out of memory" in str(e):
            print("CUDA out of memory. Releasing unused memory...")
            torch.cuda.empty_cache()
            gc.collect()
            raise e
    # # Collect the output file paths
    # del waveform1, waveform2, _
    # torch.cuda.empty_cache()
    # gc.collect()
    
    return output_paths

def cleanup_output_dir():
    save_lora_dir = "output"
    if os.path.exists(save_lora_dir):
        shutil.rmtree(save_lora_dir)
        print(f"Cleaned up directory: {save_lora_dir}")
atexit.register(cleanup_output_dir)

# Gradio interface function
def interface(audio1, audio2, alpha, num_inference_steps):
    output_paths = morph_audio_with_morphing_factor(audio1, audio2, alpha, num_inference_steps)
    return output_paths

# Gradio Interface
# demo = gr.Interface(
#     fn=interface,
#     inputs=[
#         gr.Audio(label="Upload Audio File 1", type="filepath"),
#         gr.Audio(label="Upload Audio File 2", type="filepath"),
#         gr.Slider(0, 1, step=0.01, label="Interpolation Alpha"),
#         gr.Slider(10, 50, step=1, label="Inference Steps"),
#         # gr.Textbox(label="Prompt for Audio File 1"),
#         # gr.Textbox(label="Prompt for Audio File 2"),
#     ],
#     outputs=gr.Audio(label="Interpolated Audio")
# )


with gr.Blocks() as demo:
    with gr.Tab("Sound Morphing with fixed frames."):
        gr.Markdown("### Upload two audio files for morphing")
        with gr.Row():
            audio1 = gr.Audio(label="Upload Audio File 1", type="filepath")
            audio2 = gr.Audio(label="Upload Audio File 2", type="filepath")
            num_inference_steps = gr.Slider(10, 50, step=1, label="Inference Steps", value=50)
        outputs = [
            gr.Audio(label="Morphing audio 1"),
            gr.Audio(label="Morphing audio 2"),
            gr.Audio(label="Morphing audio 3"),
            gr.Audio(label="Morphing audio 4"),
            gr.Audio(label="Morphing audio 5"),
        ]
        submit_btn1 = gr.Button("Submit")
        submit_btn1.click(morph_audio, inputs=[audio1, audio2, num_inference_steps], outputs=outputs)
    
    with gr.Tab("Sound Morphing with specified morphing factor."):
        gr.Markdown("### Upload two audio files for morphing")
        with gr.Row():
            audio1 = gr.Audio(label="Upload Audio File 1", type="filepath")
            audio2 = gr.Audio(label="Upload Audio File 2", type="filepath")
            alpha = gr.Slider(0, 1, step=0.01, label="Interpolation Alpha")
            num_inference_steps = gr.Slider(10, 50, step=1, label="Inference Steps", value=50)
        outputs=gr.Audio(label="Interpolated Audio")
        submit_btn2 = gr.Button("Submit")
        submit_btn2.click(morph_audio_with_morphing_factor, inputs=[audio1, audio2, alpha, num_inference_steps], outputs=outputs)

if __name__ == "__main__":
    demo.launch(share=True)