File size: 4,849 Bytes
18ccbe4
 
 
 
 
 
 
e085c85
18ccbe4
 
 
 
4c57756
18ccbe4
d595bbf
 
18ccbe4
d595bbf
18ccbe4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
e085c85
18ccbe4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4c57756
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
import re, io, os, stat
import tempfile, subprocess
import requests
import torch
import traceback
import numpy as np
import scipy
from importlib import import_module
from flask import Flask, Blueprint,  request, jsonify, send_file

import torch
import torchaudio

app = Flask(__name__)
# def upload_bytes(bytes, ext=".wav"):
#     return bytes

from qili import upload_bytes
# if __name__ == "__main__":
#     app = Flask(__name__)
# else:
#     app = Blueprint("xtts", __name__)


sample_root= os.environ.get('XTTS_SAMPLE_DIR') 
if(sample_root==None):
    sample_root=f'{os.getcwd()}/samples'
if not os.path.exists(sample_root):
    os.makedirs(sample_root)

default_sample=f'{os.path.dirname(os.path.abspath(__file__))}/sample.wav', f'{sample_root}/sample.pt'

ffmpeg=f'{os.path.dirname(os.path.abspath(__file__))}/ffmpeg'
try:
    st = os.stat(ffmpeg)
    os.chmod(ffmpeg, st.st_mode | stat.S_IEXEC)
except:
    traceback.print_exc() 

tts=None
model=None 
@app.route("/convert")
def predict():
    global tts 
    global model
    text = request.args.get('text')
    sample = request.args.get('sample')
    language = request.args.get('language')

    if text is None:
        return jsonify({'error': 'text is missing'}), 400
    
    text= re.sub("([^\x00-\x7F]|\w)(\.|\。|\?)",r"\1 \2\2",text)
    
    try:
        if tts is None:
            TTS=import_module("TTS.api").TTS
            model_name="tts_models/multilingual/multi-dataset/xtts_v2"
            tts = TTS(model_name=model_name)
            model=tts.synthesizer.tts_model
            #hack to use cache
            model.__get_conditioning_latents=model.get_conditioning_latents
            model.get_conditioning_latents=get_conditioning_latents

        wav = tts.tts(
            text, 
            language=language if language is not None else "zh",
            speaker_wav=sample if sample is not None else default_sample[0],
        )

        with io.BytesIO() as wav_buffer:
            if torch.is_tensor(wav):
                wav = wav.cpu().numpy()
            if isinstance(wav, list):
                wav = np.array(wav)
            wav_norm = wav * (32767 / max(0.01, np.max(np.abs(wav))))
            wav_norm = wav_norm.astype(np.int16)
            scipy.io.wavfile.write(wav_buffer, tts.synthesizer.output_sample_rate, wav_norm)
            wav_bytes = wav_buffer.getvalue()
            url= upload_bytes(wav_bytes, ext=".wav")
            print(f'wav is at {url}')
            return url
    except Exception as e:
        traceback.print_exc()
        return str(e)

@app.route("/play")
def play():
    url=predict()
    return f'''
        <html>
            <body>
                <audio controls autoplay>
                    <source src="{url}" type="audio/wav">
                    Your browser does not support the audio element.
                </audio>
            </body>
        </html>
    '''

def get_conditioning_latents(audio_path, **others):
    global model
    speaker_wav, pt_file=download(audio_path)
    try:
        if pt_file != None:
            (
                gpt_cond_latent,
                speaker_embedding,
            ) = torch.load(pt_file)
            print(f'sample wav info loaded from {pt_file}')
    except:
        (
            gpt_cond_latent,
            speaker_embedding,
        ) = model.__get_conditioning_latents(audio_path=speaker_wav, **others)
        torch.save((gpt_cond_latent,speaker_embedding), pt_file)
        print(f'sample wav info saved to {pt_file}')
    return gpt_cond_latent,speaker_embedding

def download(url):
    try:
        response = requests.get(url)
        if response.status_code == 200:
            id=f'{sample_root}/{response.headers["etag"]}.pt'.replace('"','')
            if(os.path.exists(id)):
                return "", id
            with tempfile.NamedTemporaryFile(mode="wb", delete=True) as temp_file:
                temp_file.write(response.content)
                return trim_sample_audio(os.path.abspath(temp_file.name)), id
    except:
        return default_sample

def trim_sample_audio(speaker_wav):
    global ffmpeg
    try:
        lowpass_highpass = "lowpass=8000,highpass=75,"
        trim_silence = "areverse,silenceremove=start_periods=1:start_silence=0:start_threshold=0.02,areverse,silenceremove=start_periods=1:start_silence=0:start_threshold=0.02,"
        out_filename=speaker_wav.replace(".wav","_trimed.wav")
        shell_command = f"{ffmpeg} -y -i {speaker_wav} -af {lowpass_highpass}{trim_silence} {out_filename}".split(" ")
        subprocess.run(
            [item for item in shell_command],
            capture_output=False,
            text=True,
            check=True,
        )
        return out_filename
    except:
        traceback.print_exc()
        return speaker_wav


@app.route("/")
def hello():
    return "hello xtts"