File size: 7,059 Bytes
abf6bf2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
import runpod
import requests
from voice_generation import generate_wav
import boto3
import os
import uuid
from pydub import AudioSegment
import time
from typing import Optional


AWS_ACCESS_KEY_ID = os.environ.get('AWS_ACCESS_KEY_ID')
AWS_SECRET_ACCESS_KEY = os.environ.get('AWS_SECRET_ACCESS_KEY')


models = {
    'kanye': 'weights/kanye.pth',
    'rose-bp': 'weights/rose-bp.pth',
    'jungkook': 'weights/jungkook.pth',
    'iu': 'weights/iu.pth',
    'drake': 'weights/drake.pth',
    'ariana-grande': 'weights/ariana-grande.pth'
}


print('run handler. Removed 2nd gen')


def combine_audio(voice_path: str, instrumental_path: str, optional_path: Optional[str] = None):
    audio1 = AudioSegment.from_file(instrumental_path)
    audio2 = AudioSegment.from_file(voice_path)
    
    length = max(len(audio1), len(audio2))
    audio1 = audio1 + AudioSegment.silent(duration=length - len(audio1))
    audio2 = audio2 + AudioSegment.silent(duration=length - len(audio2))
    
    if optional_path:
        audio3 = AudioSegment.from_file(optional_path)
        audio3 = audio3 + AudioSegment.silent(duration=length - len(audio3))
        combined = audio1.overlay(audio2).overlay(audio3)
    else:
        combined = audio1.overlay(audio2)
    
    combined.export("combined.mp3", format="mp3")


def upload_file_to_s3(local_file_path, s3_file_path):
    bucket_name = 'voice-gen-audios'
    s3 = boto3.client('s3', aws_access_key_id=AWS_ACCESS_KEY_ID, aws_secret_access_key=AWS_SECRET_ACCESS_KEY)
    try:
        s3.upload_file(local_file_path, bucket_name, s3_file_path)
        return {"url": f"https://{bucket_name}.s3.eu-north-1.amazonaws.com/{s3_file_path}"}
    except boto3.exceptions.S3UploadFailedError as e:
        return {"error": f"failed to upload file {local_file_path} to s3 as {s3_file_path}"}


def clean_up_files(remove_voice_model=False, remove_back_vocal=False):
    files = [
        "instrumental.mp3",
        "vocal.mp3",
        "output_voice.wav",
        "combined.mp3",
    ]
    if remove_voice_model:
        files.append("voice_model.pth")
    if remove_back_vocal:
        files.append("back_vocal.mp3")
    for file in files:
        try:
            os.remove(file)
        except FileNotFoundError:
            return {"error": f"failed to remove file {file}"}
    return {"success": "files removed successfully"}


def get_voice_model(event):
    voice_model_id = event["input"].get("voice_model_id", "")
    voice_model_url = event["input"].get("voice_model_url", "")
    
    if not voice_model_url and not voice_model_id:
        return {"error": "voice_model_url or voice_model_id is required"}

    if voice_model_id and voice_model_id not in models:
        return {"error": "model not found in pre-loaded models"}
    
    if voice_model_id:
        return {"model_path": models[voice_model_id]}
    
    print("downloading voice_model")
    voice_model_response = requests.get(voice_model_url)
    if voice_model_response.status_code != 200:
        return {"error": f"failed to download voice_model, error: {voice_model_response.text}"}
    
    with open("voice_model.pth", "wb") as f:
        f.write(voice_model_response.content)

    return {"model_path": "voice_model.pth"}


def get_method(event):
    method = event["input"].get("method", "pm")
    if method not in ["pm", "harvest"]:
        method = "pm"
    return method


def get_index_rate(event):
    index_rate = event["input"].get("index_rate", 0.6)
    if index_rate < 0 or index_rate > 1:
        index_rate = 0.6
    return index_rate


def handler(event):
    print(event)
    file_id = str(uuid.uuid4())
    user_id = event["input"].get("user_id", "not provided")
    
    if not AWS_ACCESS_KEY_ID or not AWS_SECRET_ACCESS_KEY:
        return {"error": "AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY are missing from environment variables"}
    
    voice_model = get_voice_model(event)
    if "error" in voice_model:
        return voice_model.get("error")

    vocal_url = event["input"].get("vocal_url", "")
    if vocal_url == "":
        return {"error": "vocal_url is required"}
    
    instrumental_url = event["input"].get("instrumental_url", "")
    if instrumental_url == "":
        return {"error": "instrumental_url is required"}
    
    vocal_file = requests.get(vocal_url)
    if vocal_file.status_code != 200:
        return {"error": "failed to download vocal_file"}
    with open("vocal.mp3", "wb") as f:
        f.write(vocal_file.content)
    
    instrumental_file = requests.get(instrumental_url)
    if instrumental_file.status_code != 200:
        return {"error": "failed to download instrumental_file"}
    with open("instrumental.mp3", "wb") as f:
        f.write(instrumental_file.content)

    back_vocal_url = event["input"].get("back_vocal_url", "")
    if back_vocal_url != "":
        back_vocal_file = requests.get(back_vocal_url)
        if back_vocal_file.status_code != 200:
            return {"error": "failed to download back_vocal_file"}
        with open("back_vocal.mp3", "wb") as f:
            f.write(back_vocal_file.content)

    generation_start = time.time()  # remove after testing

    generation = generate_wav(
        audio_file='vocal.mp3',
        method=get_method(event),
        index_rate=get_index_rate(event),
        output_file='output_voice.wav',
        model_path=voice_model.get("model_path")
    )
    generation_end = time.time()  # remove after testing
    time_taken_generation = generation_end - generation_start  # remove after testing
    print(f"generation took {time_taken_generation} seconds")  # remove after testing
    if "error" in generation:
        return generation.get("error")

    if back_vocal_url:
        combine_audio("output_voice.wav", "instrumental.mp3", "back_vocal.mp3")
    else:
        combine_audio("output_voice.wav", "instrumental.mp3")

    if not os.path.exists("combined.mp3"):
        return {"error": "failed to combine audio"}

    combined = upload_file_to_s3("combined.mp3", f"{file_id}.mp3")
    output_voice = upload_file_to_s3("output_voice.wav", f"{file_id}-generated-voice.wav")

    if combined_error := combined.get("error"):
        return combined_error
    
    if output_voice_error := output_voice.get("error"):
        return output_voice_error
    
    combined_url = combined.get("url")
    output_voice_url = output_voice.get("url")

    need_to_remove_voice_model = False
    need_to_remove_back_vocal = False
    if voice_model.get("model_path") == "voice_model.pth":
        need_to_remove_voice_model = True
    if back_vocal_url:
        need_to_remove_back_vocal = True
    cleanup_result = clean_up_files(need_to_remove_voice_model, need_to_remove_back_vocal)
    if cleanup_error := cleanup_result.get("error"):
        return cleanup_error

    return {
        "combined_url": combined_url,
        "output_voice_url": output_voice_url,
        "user_id": user_id,
        "time_taken_generation": time_taken_generation,  # remove after testing
    }


runpod.serverless.start({"handler": handler})