add flask

demo/serverFlask.py

@@ -0,0 +1,136 @@
+from flask import Flask, request, Markup, abort, jsonify
+from flask_cors import CORS
+import logging
+from logging.config import dictConfig
+import sys
+import base64
+
+import torch
+import numpy as np
+from scipy.io.wavfile import write, read
+from datetime import datetime
+
+import traceback
+import struct
+
+sys.path.append("mod")
+sys.path.append("mod/text")
+
+import utils
+from data_utils import TextAudioSpeakerLoader, TextAudioSpeakerCollate
+from models import SynthesizerTrn
+from text.symbols import symbols
+
+dictConfig({
+    'version': 1,
+    'formatters': {'default': {
+        'format': '[%(asctime)s] %(levelname)s in %(module)s: %(message)s',
+    }},
+    'handlers': {'wsgi': {
+        'class': 'logging.StreamHandler',
+        'stream': 'ext://flask.logging.wsgi_errors_stream',
+        'formatter': 'default'
+    }},
+    'root': {
+        'level': 'INFO',
+        'handlers': ['wsgi']
+    }
+})
+
+app = Flask(__name__, static_folder="/voice-changer-internal/frontend/dist", static_url_path='/')
+CORS(app, resources={r"/*": {"origins": "*"}}) 
+
+class VoiceChanger():
+    def __init__(self, config, model):
+        self.hps =utils.get_hparams_from_file(config)
+        self.net_g = SynthesizerTrn(
+                len(symbols),
+                self.hps.data.filter_length // 2 + 1,
+                self.hps.train.segment_size // self.hps.data.hop_length,
+                n_speakers=self.hps.data.n_speakers,
+                **self.hps.model)
+        self.net_g.eval()
+        self.gpu_num = torch.cuda.device_count()
+        print("GPU_NUM:",self.gpu_num)
+        utils.load_checkpoint( model, self.net_g, None)
+
+
+    def on_request(self, gpu, srcId, dstId, timestamp, wav): 
+        if wav==0:
+            samplerate, data=read("dummy.wav")
+            unpackedData = data
+        else:
+            unpackedData = np.array(struct.unpack('<%sh'%(len(wav) // struct.calcsize('<h') ), wav))
+            write("logs/received_data.wav", 24000, unpackedData.astype(np.int16))
+
+        try:
+            if gpu<0 or self.gpu_num==0 :
+                with torch.no_grad():
+                    dataset = TextAudioSpeakerLoader("dummy.txt", self.hps.data, no_use_textfile=True)
+                    data = dataset.get_audio_text_speaker_pair([ unpackedData, srcId, "a"])
+                    data = TextAudioSpeakerCollate()([data])
+                    x, x_lengths, spec, spec_lengths, y, y_lengths, sid_src = [x.cpu() for x in data]
+                    sid_tgt1 = torch.LongTensor([dstId]).cpu()
+                    audio1 = (self.net_g.cpu().voice_conversion(spec, spec_lengths, sid_src=sid_src, sid_tgt=sid_tgt1)[0][0,0].data * self.hps.data.max_wav_value).cpu().float().numpy()
+            else:
+                with torch.no_grad():
+                    dataset = TextAudioSpeakerLoader("dummy.txt", self.hps.data, no_use_textfile=True)
+                    data = dataset.get_audio_text_speaker_pair([ unpackedData, srcId, "a"])
+                    data = TextAudioSpeakerCollate()([data])
+                    x, x_lengths, spec, spec_lengths, y, y_lengths, sid_src = [x.cuda(gpu) for x in data]
+                    sid_tgt1 = torch.LongTensor([dstId]).cuda(gpu)
+                    audio1 = (self.net_g.cuda(gpu).voice_conversion(spec, spec_lengths, sid_src=sid_src, sid_tgt=sid_tgt1)[0][0,0].data * self.hps.data.max_wav_value).cpu().float().numpy()
+        except Exception as e:
+            print("VC PROCESSING!!!! EXCEPTION!!!", e)
+            print(traceback.format_exc())
+        
+        audio1 = audio1.astype(np.int16)
+        return audio1
+
+
+
+@app.route('/test', methods=['GET', 'POST'])
+def test():
+    try:
+        if request.method == 'GET':
+            return request.args.get('query', '')
+        elif request.method == 'POST':
+            print("POST REQUEST PROCESSING....")
+            gpu = int(request.json['gpu'])
+            srcId = int(request.json['srcId'])
+            dstId = int(request.json['dstId'])
+            timestamp = int(request.json['timestamp'])
+            buffer = request.json['buffer']
+            wav = base64.b64decode(buffer)
+            # print(wav)
+            # print(base64.b64encode(wav))
+            changedVoice = voiceChanger.on_request(gpu, srcId, dstId, timestamp, wav)
+            changedVoiceBase64 = base64.b64encode(changedVoice).decode('utf-8')
+            # print("changedVoice",changedVoice)
+            # print("CV64",changedVoiceBase64)
+            data = {
+                "gpu":gpu,
+                "srcId":srcId,
+                "dstId":dstId,
+                "timestamp":timestamp,
+                "changedVoiceBase64":changedVoiceBase64
+            }
+            return jsonify(data)
+        else:
+            return abort(400)
+    except Exception as e:
+        print("REQUEST PROCESSING!!!! EXCEPTION!!!", e)
+        print(traceback.format_exc())
+        return str(e)
+
+
+if __name__ == '__main__':
+    args = sys.argv
+    PORT = args[1]
+    CONFIG = args[2]
+    MODEL  = args[3]
+    app.logger.info('INITIALIZE MODEL')
+    voiceChanger = VoiceChanger(CONFIG, MODEL)
+    voiceChanger.on_request(0,0,0,0,0)
+    app.logger.info('START APP')
+    app.run(debug=True, host='0.0.0.0',port=PORT)

demo/serverSIO.py

@@ -17,7 +17,7 @@ from text.symbols import symbols
 
 
 
-class MyCustomNamespace(socketio.Namespace): # 名前空間を設定するクラス
+class MyCustomNamespace(socketio.Namespace): 
     def __init__(self, namespace, config, model):
         super().__init__(namespace)
         self.hps =utils.get_hparams_from_file(config)
@@ -36,7 +36,7 @@ class MyCustomNamespace(socketio.Namespace): # 名前空間を設定するクラ
         print('[{}] connet sid : {}'.format(datetime.now().strftime('%Y-%m-%d %H:%M:%S') , sid))
         # print('[{}] connet env : {}'.format(datetime.now().strftime('%Y-%m-%d %H:%M:%S') , environ))
 
-    def on_request_message(self, sid, msg): # 送信してきたクライアントだけにメッセージを送る関数
+    def on_request_message(self, sid, msg): 
         # print("MESSGaa", msg)
         gpu = int(msg[0])
         srcId = int(msg[1])
@@ -88,9 +88,9 @@ if __name__ == '__main__':
     print(f"start... PORT:{PORT}, CONFIG:{CONFIG}, MODEL:{MODEL}")    
     # sio = socketio.Server(cors_allowed_origins='http://localhost:8080') 
     sio = socketio.Server(cors_allowed_origins='*') 
-    sio.register_namespace(MyCustomNamespace('/test', CONFIG, MODEL)) # 名前空間を設定
+    sio.register_namespace(MyCustomNamespace('/test', CONFIG, MODEL)) 
     app = socketio.WSGIApp(sio,static_files={
         '': '../frontend/dist',
-    }) # wsgiサーバーミドルウェア生成
+    }) 
-    eventlet.wsgi.server(eventlet.listen(('0.0.0.0',int(PORT))), app) # wsgiサーバー起動
+    eventlet.wsgi.server(eventlet.listen(('0.0.0.0',int(PORT))), app) 
     

demo/setupFlask.sh

@@ -0,0 +1,14 @@
+#!/bin/bash
+
+echo config: $1
+echo model: $2
+cp -r /resources/* .
+
+if [[ -e ./setting.json ]]; then
+  cp ./setting.json ../frontend/dist/assets/setting.json
+fi
+pip install flask
+pip install flask_cors
+python3 serverFlask.py 8080 $1 $2
+
+

docs/assets/setting.json

@@ -1,5 +1,6 @@
 {
     "app_title": "voice-changer",
+    "majar_mode": "docker",
     "voice_changer_server_url": "http://localhost:8080/test",
     "sample_rate": 48000,
     "buffer_size": 1024,

start2.sh

@@ -2,7 +2,7 @@
 
 # 参考:https://programwiz.org/2022/03/22/how-to-write-shell-script-for-option-parsing/
 
-DOCKER_IMAGE=dannadori/voice-changer:20220829_195844
+DOCKER_IMAGE=dannadori/voice-changer:20220831_141509
 TENSORBOARD_PORT=6006
 VOICE_CHANGER_PORT=8080
 

trainer/Dockerfile

@@ -1,4 +1,4 @@
-FROM dannadori/voice-changer-internal:20220829_195653 as front
+FROM dannadori/voice-changer-internal:20220831_140605 as front
 FROM debian:bullseye-slim as base
 
 ARG DEBIAN_FRONTEND=noninteractive