Spaces:
Running
Running
File size: 3,386 Bytes
802e11f 5293275 802e11f 9c7fd5c 3ebe5da 802e11f b5433b4 802e11f b5433b4 802e11f b5433b4 802e11f 5293275 802e11f b5433b4 802e11f 0b2bb7e 802e11f 0b2bb7e 802e11f b5433b4 802e11f 67f7fe4 3ebe5da 9079195 802e11f b5433b4 802e11f |
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 |
"""
Copyright 2023 Balacoon
contains implementation
for voice conversion request
"""
import os
import asyncio
import base64
import hashlib
import json
import ssl
import time
from typing import Tuple
import numpy as np
import resampy
import websockets
def prepare_audio(audio: Tuple[int, np.ndarray]) -> np.ndarray:
"""
ensures that audio is in int16 format, 16khz mono
"""
sr, wav = audio
# ensure proper type
if wav.dtype == np.int32:
max_val = np.max(np.abs(wav))
mult = (32767.0 / 2**31) if max_val > 32768 else 1.0
wav = (wav.astype(np.float32) * mult).astype(np.int16)
elif wav.dtype == np.float32 or wav.dtype == np.float64:
mult = 32767.0 if np.max(np.abs(wav)) <= 1.0 else 1.0
wav = (wav * mult).astype(np.int16)
if wav.ndim == 2:
# average channels
if wav.shape[0] == 2:
wav = np.mean(wav, axis=0, keepdims=False)
if wav.shape[1] == 2:
wav = np.mean(wav, axis=1, keepdims=False)
if wav.ndim != 1:
return None
# ensure proper sampling rate
if sr != 16000:
wav = (wav / 32768.0).astype(np.float)
wav = resampy.resample(wav, sr, 16000)
wav = (wav * 32768.0).astype(np.int16)
return wav
def create_signature(api_secret: str) -> str:
"""
helper function that creates signature,
required to authentificate the request
"""
int_time = int(time.time() / 1000)
signature_input = (api_secret + str(int_time)).encode()
signature = hashlib.sha256(signature_input).hexdigest()
return signature
async def async_service_request(source: np.ndarray, target: np.ndarray, api_key: str, api_secret: str) -> np.ndarray:
ssl_context = ssl.create_default_context()
async with websockets.connect(
os.environ["endpoint"], close_timeout=1024, ssl=ssl_context
) as websocket:
request_dict = {
"source": base64.b64encode(source.tobytes()).decode("utf-8"),
"target": base64.b64encode(target.tobytes()).decode("utf-8"),
"api_key": api_key,
"signature": create_signature(api_secret),
}
request = json.dumps(request_dict)
await websocket.send(request)
# read reply
result_lst = []
while True:
try:
data = await asyncio.wait_for(websocket.recv(), timeout=15)
result_lst.append(np.frombuffer(data, dtype="int16"))
except websockets.exceptions.ConnectionClosed:
break
except asyncio.TimeoutError:
break
if data is None:
break
result = np.concatenate(result_lst) if result_lst else None
return result
def vc_service_request(
source_audio: Tuple[int, np.ndarray], target_audio: Tuple[int, np.ndarray],
api_key: str, api_secret: str,
) -> Tuple[int, np.ndarray]:
"""
prepares audio (has to be 16khz mono)
and runs request to a voice conversion service
"""
src = prepare_audio(source_audio)
tgt = prepare_audio(target_audio)
if src is None or tgt is None:
return
if len(src) >= 60 * 16000 or len(tgt) >= 30 * 16000:
# input is way too long, dont return anything
return
res = asyncio.run(async_service_request(src, tgt, api_key, api_secret))
return 16000, res
|