first commit

README.md

@@ -0,0 +1,3 @@
+# asr_serve
+
+a local deploy ASR models web server

app/asr.py

@@ -0,0 +1,40 @@
+import os, sys
+import argparse
+from appPublic.log import MyLogger, info, debug, warning
+from appPublic.folderUtils import ProgramPath
+from appPublic.jsonConfig import getConfig
+from appPublic.registerfunction import RegisterFunction
+from ahserver.configuredServer import ConfiguredServer
+from ahserver.serverenv import ServerEnv
+from whisper_model import WhisperFile
+from asr_engine import generate
+
+__version__ = '0.0.1'
+
+if __name__ == '__main__':
+	parser = argparse.ArgumentParser(prog="Sage")
+	parser.add_argument('-w', '--workdir')
+	parser.add_argument('-p', '--port')
+	args = parser.parse_args()
+	print(args)
+	workdir = args.workdir or os.getcwd()
+	p = ProgramPath()
+	config = getConfig(workdir, NS={'workdir':workdir, 'ProgramPath':p})
+	if config.logger:
+		logger = MyLogger(config.logger.name or 'sage',
+							levelname=config.logger.levelname or 'debug',
+							logfile=config.logger.logfile or None)
+	else:
+		logger = MyLogger('sage', levelname='debug')
+
+	info(f'========sage version={__version__}========')
+	# server = ConfiguredServer(auth_klass=MyAuthAPI, workdir=workdir)
+	server = ConfiguredServer(workdir=workdir)
+	rf = RegisterFunction()
+	rf.register('generate', generate)
+	g = ServerEnv()
+	g.whisper_engine = WhisperFile()
+	port = args.port or config.website.port or 8080
+	port = int(port)
+	server.run(port=port)
+

app/asr_engine.py

@@ -0,0 +1,76 @@
+import os
+import time
+from traceback import print_exc
+import base64
+import wave
+from appPublic.log import info, debug, warning, error, exception, critical
+from appPublic.dictObject import DictObject
+from appPublic.folderUtils import temp_file
+from ahserver.serverenv import ServerEnv
+from aiohttp.web import StreamResponse
+from ahserver.globalEnv import realpath
+
+def save_base64_wav(base64_data, output_file,sample_rate=16000, num_channels=1):
+    # Decode the base64 data
+    wav_data = base64.b64decode(base64_data)
+
+    # Open a new WAV file for writing
+    with wave.open(output_file, 'wb') as wf:
+        # Set the parameters of the WAV file
+        wf.setnchannels(num_channels)  # Mono channel
+        wf.setsampwidth(2)  # 16-bit sample width
+        wf.setframerate(sample_rate)  # 44.1 kHz sample rate
+        # Write the decoded data to the WAV file
+        wf.writeframes(wav_data)
+
+async def generate(request, kw):
+	params_kw = kw.get('params_kw', DictObject())
+	info(f'{params_kw=}')
+	model = params_kw.model
+	audio_file = params_kw.audio_file
+	if not audio_file:
+		audio = params_kw.audio
+		if audio is None:
+			return {
+				'status':'error',
+				'message':'audio is null'
+			}
+		audio_file = temp_file(suffix='.wav')
+		save_base64_wav(audio, audio_file)
+	else:
+		audio_file = realpath(audio_file)
+
+	engine = None
+	g = ServerEnv()
+	if model=='whisper':
+		engine = g.whisper_engine
+
+	if engine is None:
+		return {
+			'status':'error',
+			'message':f'model={model} is not defined'
+		}
+	try:
+		t1 = time.time()
+		dic = await engine.stt(audio_file)
+		t2 = time.time()
+		os.remove(audio_file)
+		ret = {
+			"status":"ok",
+			"time_cost":t2-t1,
+			"content":dic['text'],
+			"segments":dic['segments'],
+			"language":dic['language']
+		}
+		info(f'{dic=}, {ret=}')
+		return ret
+	except Exception as e:
+		exception(f'{e}')
+		print_exc()
+		return {
+			'status':'error',
+			'message':f'{e}'
+		}
+
+g = ServerEnv()
+g.generate = generate

app/sensevoice/load_model.py

@@ -0,0 +1,22 @@
+from ahserver.serverenv import ServerEnv
+from appPublic.worker import awaitify
+from appPublic.jsonConfig import getConfig, get_definition
+from funasr import AutoModel
+from funasr.utils.postprocess_utils import rich_transcription_postprocess
+
+class SenseVoiceBase:
+    def __init__(self):
+        model_dir = get_definition('sensevoice_path')
+        self.model = AutoModel(model=model_dir)
+    
+    def _stt(self, x):
+        pass
+    stt = awaitify(_stt)
+
+class SenseVoiceFile(SenseVoiceBase):
+    def _stt(self, filepath):
+        res = this.model.generate(input=filepath,language="auto",
+                use_itn=True,
+                batch_size_s=60
+        )
+        return rich_transcription_postprocess(res[0]["text"])

app/whisper_model.py

@@ -0,0 +1,54 @@
+from ahserver.serverenv import ServerEnv
+from appPublic.worker import awaitify
+from appPublic.log import info, debug, warning, error, exception, critical
+
+from ahserver.globalEnv import get_definition
+import numpy as np
+import base64
+import whisper
+
+# 编码
+def base64_encode(text):
+	text_bytes = text.encode('utf-8')
+	encoded_bytes = base64.b64encode(text_bytes)
+	encoded_text = encoded_bytes.decode('utf-8')
+	return encoded_text
+
+# 解码
+def base64_decode(encoded_text):
+	encoded_bytes = encoded_text.encode('utf-8')
+	decoded_bytes = base64.b64decode(encoded_bytes)
+	decoded_text = decoded_bytes.decode('utf-8')
+	return decoded_text
+
+class WhisperBase:
+	def __init__(self):
+		model_name = get_definition('whisper_model')
+		self.model = whisper.load_model(model_name)
+
+	def _stt(self, filepath):
+		e = Exception(f'{filepath=} WhisperBase can not use')
+		exception(f'{e=}')
+		raise e
+
+
+class WhisperFile(WhisperBase):
+	def _stt(self, filepath):
+		return self.model.transcribe(filepath)
+	
+	stt = awaitify(_stt)
+
+class WhisperBase64(WhisperBase):
+	def _stt(self, audio):
+		audiolist = audio.values()
+		nparr = np.array(audiolist, dtype=np.float32)
+		nparr=np.vstack(nparr).astype(np.float)
+		"""
+		
+		raw = base64_decode(audio_base64)
+		ndarr = np.frombuffer(raw, dtype=np.float32)
+		"""
+		info(f'ndarr={nparr}')
+		return self.model.transcribe(nparr)
+
+	stt = awaitify(_stt)

conf/config.json

@@ -0,0 +1,86 @@
+{
+	"logger":{
+		"name":"asr",
+		"levelname":"info",
+		"logfile":"$[workdir]$/logs/asr.log"
+	},
+	"definitions":{
+		"whisper_model":"/d/ymq/osc/models/whisper/large-v3.pt"
+	},
+	"filesroot":"$[workdir]$/files",
+	"databases":{
+		"sage":{
+			"driver":"aiomysql",
+			"async_mode":true,
+			"coding":"utf8",
+			"maxconn":100,
+			"dbname":"sage",
+			"kwargs":{
+				"user":"test",
+				"db":"sage",
+				"password":"QUZVcXg5V1p1STMybG5Ia6mX9D0v7+g=",
+				"host":"localhost"
+			}
+		}
+	},
+	"website":{
+		"paths":[
+			["$[workdir]$/wwwroot",""]
+		],
+		"client_max_size":10000,
+		"host":"0.0.0.0",
+		"port":9091,
+		"coding":"utf-8",
+		"ssl_gg":{
+			"crtfile":"$[workdir]$/conf/www.bsppo.com.pem",
+			"keyfile":"$[workdir]$/conf/www.bsppo.com.key"
+		},
+		"indexes":[
+			"index.html",
+			"index.tmpl",
+			"index.ui",
+			"index.dspy",
+			"index.md"
+		],
+		"startswiths":[
+			{
+				"leading":"/generate",
+				"registerfunction":"generate"
+			}
+		],
+		"processors":[
+			[".ws","ws"],
+			[".xterm","xterm"],
+			[".proxy","proxy"],
+			[".llm", "llm"],
+			[".llms", "llms"],
+			[".llma", "llma"],
+			[".xlsxds","xlsxds"],
+			[".sqlds","sqlds"],
+			[".tmpl.js","tmpl"],
+			[".tmpl.css","tmpl"],
+			[".html.tmpl","tmpl"],
+			[".bcrud", "bricks_crud"],
+			[".tmpl","tmpl"],
+			[".bui","bui"],
+			[".ui","bui"],
+			[".dspy","dspy"],
+			[".md","md"]
+		],
+		"rsakey":{
+			"privatekey":"$[workdir]$/conf/rsa_private_key.pem",
+			"publickey":"$[workdir]$/conf/rsa_public_key.pem"
+		},
+		"session_max_time":3000,
+		"session_issue_time":2500,
+		"session_redis_oops":{
+			"url":"redis://127.0.0.1:6379"
+		}
+	},
+	"langMapping":{
+		"zh-Hans-CN":"zh-cn",
+		"zh-CN":"zh-cn",
+		"en-us":"en",
+		"en-US":"en"
+	}
+}

requirements.txt

@@ -0,0 +1,2 @@
+openai-whisper
+