first commit

README.md

@@ -0,0 +1,35 @@
+# faster-whisper
+这个比openai-whisper要快
+
+## 创建虚拟环境
+```
+mkdir ve
+python3 -m venv ~/ve/fw
+```
+
+## 安装
+
+```
+source ~/ve/fw/bin/activate
+pip install nvidia-cublas-cu12 nvidia-cudnn-cu12
+
+export LD_LIBRARY_PATH=`python -c 'import os; import nvidia.cublas.lib; import nvidia.cudnn.lib; print(os.path.dirname(nvidia.cublas.lib.__file__) + ":" + os.path.dirname(nvidia.cudnn.lib.__file__))'`
+pip install faster-whisper
+```
+
+## 运行
+创建一个run.sh
+```
+#!/bin/sh
+
+r=$HOME/ve/fw/bin/python
+export LD_LIBRARY_PATH=`$r -c 'import os; import nvidia.cublas.lib; import nvidia.cudnn.lib; print(os.path.dirname(nvidia.cublas.lib.__file__) + ":" + os.path.dirname(nvidia.cudnn.lib.__file__))'`
+$r $*
+```
+
+## 执行服务
+./run.sh zmqfw.py
+
+## 执行客户端
+./run.shzmq_client.py
+

conf/config.json

@@ -0,0 +1,6 @@
+{
+	"zmq_url" : "tcp://127.0.0.1:10002",
+	"model_dir" : "large-v3",
+	"device" : "cuda",
+	"conpute_type": "float16"
+}

run.sh

@@ -0,0 +1,5 @@
+#!/bin/sh
+
+export LD_LIBRARY_PATH=/data/ymq/ve/fw/lib/python3.12/site-packages/nvidia/cublas/lib:/data/ymq/ve/fw/lib/python3.12/site-packages/nvidia/cudnn/lib
+r=$HOME/ve/fw/bin/python
+$r $*

zmq_client.py

@@ -0,0 +1,42 @@
+import json
+import os
+
+from appPublic.dictObject import DictObject
+from appPublic.zmq_reqrep import ZmqRequester
+from appPublic.jsonConfig import getConfig
+
+zmq_url = "tcp://127.0.0.1:9999"
+from time import time
+
+class ASRClient:
+    def __init__(self, zmq_url):
+        self.zmq_url = zmq_url
+        self.requester = ZmqRequester(self.zmq_url)
+
+    def generate(self, audio_file):
+        d = {
+            "audio_file":audio_file
+        }
+        msg = json.dumps(d)
+        print(f'send={msg}')
+        resp = self.requester.send(msg)
+        if resp != None:
+            ret = json.loads(resp)
+            print(f'response={ret}')
+        else:
+            print(f'response is None')
+
+    def run(self):
+        print(f'running {self.zmq_url}')
+        while True:
+            print('input audio_file:')
+            af = input()
+            if len(af) > 0:
+                self.generate(af)
+        print('ended ...')
+
+if __name__ == '__main__':
+    workdir = os.getcwd()
+    config = getConfig(workdir)
+    asr = ASRClient(config.zmq_url or zmq_url)
+    asr.run()

zmqfw.py

@@ -0,0 +1,57 @@
+import json
+import os
+
+from appPublic.dictObject import DictObject
+from appPublic.zmq_reqrep import ZmqReplier
+from appPublic.jsonConfig import getConfig
+
+from time import time
+from faster_whisper import WhisperModel
+
+class ZmqASR:
+    def __init__(self):
+        config = getConfig()
+        self.model_dir = config.model_dir or "large-v3"
+        self.device = config.device or "cpu"
+        self.zmq_url = config.zmq_url
+        self.compute_type=config.compute_type or "float16"
+        self.model = WhisperModel(self.model_dir, 
+                                        device=self.device, 
+                                        compute_type=self.compute_type)
+        self.replier = ZmqReplier(self.zmq_url, self.generate)
+
+    def run(self):
+        print(f'running {self.zmq_url}')
+        self.replier._run()
+        print('ended ...')
+
+    def generate(self, d):
+        msg= d.decode('utf-8')
+        data = DictObject(**json.loads(msg))
+        t1 = time()
+        segs, info = self.model.transcribe(data.audio_file)
+        t2 = time()
+        d = DictObject(**{
+            'segments':[],
+            'language':info.language
+        })
+        d.content = ''
+        for s in segs:
+            seg = {
+                "start":s.start,
+                "end":s.end,
+                "text":s.text
+            }
+            d.segments.append(seg)
+            d.content += s.text
+        d.time_cost = t2 - t1
+        print(f'{d}')
+        return json.dumps(d)
+
+if __name__ == '__main__':
+    workdir = os.getcwd()
+    config = getConfig(workdir)
+    print(f'{config=}')
+    asr = ZmqASR()
+    print('here')
+    asr.run()