Merge pull request #13 from makaveli10/update_faster_whisper

Update faster whisper.

Audio-Transcription-Firefox/content.js

@@ -45,8 +45,12 @@ function startRecording() {
       socket.send("handshake");
     };
 
+    let isServerReady = false;
     socket.onmessage = (event) => {
-      // console.log(event.data);
+      if (!isServerReady){
+        isServerReady = true;
+        return;
+      }
       const data = event.data;
       browser.runtime.sendMessage({ action: "transcript", data })
           .catch(function(error) {
@@ -64,7 +68,7 @@ function startRecording() {
       recorder = audioContext.createScriptProcessor(4096, 1, 1);
 
       recorder.onaudioprocess = async (event) => {
-        if (!audioContext || !isCapturing) return;
+        if (!audioContext || !isCapturing || !isServerReady) return;
 
         const inputData = event.inputBuffer.getChannelData(0);
         const audioData16kHz = resampleTo16kHZ(inputData, audioContext.sampleRate);

Audio-Transcription/options.js

@@ -81,25 +81,32 @@ async function startRecord(option) {
     };
 
     const socket = new WebSocket("ws://localhost:9090/");
+    let isServerReady = false;
     socket.onopen = function(e) { 
       socket.send("handshake");
     };
 
     socket.onmessage = async (event) => {
-      // console.log(event.data);
+      console.log(event.data);
+      if (isServerReady === false){
+        isServerReady = true;
+        return;
+      }
+      
       res = await sendMessageToTab(option.currentTabId, {
         type: "transcript",
         data: event.data,
       });
     };
 
+    
     const audioDataCache = [];
     const context = new AudioContext();
     const mediaStream = context.createMediaStreamSource(stream);
     const recorder = context.createScriptProcessor(4096, 1, 1);
 
     recorder.onaudioprocess = async (event) => {
-      if (!context) return;
+      if (!context || !isServerReady) return;
 
       const inputData = event.inputBuffer.getChannelData(0);
       const audioData16kHz = resampleTo16kHZ(inputData, context.sampleRate);
@@ -114,6 +121,7 @@ async function startRecord(option) {
     mediaStream.connect(recorder);
     recorder.connect(context.destination);
     mediaStream.connect(context.destination);
+    // }
   } else {
     window.close();
   }

client.py

@@ -17,11 +17,15 @@ FORMAT = pyaudio.paInt16
 CHANNELS = 1
 RATE = 16000
 RECORD_SECONDS = 60000
-
+START_RECORDING = False
 
 
 def on_message(ws, message):
+    global START_RECORDING
     message = json.loads(message)
+    if message == "SERVER_READY":
+        START_RECORDING = True
+        return
     text = []
     if len(message):
         for seg in message:
@@ -35,6 +39,9 @@ def on_message(ws, message):
     wrapper = textwrap.TextWrapper(width=60)
     word_list = wrapper.wrap(text="".join(text))
     # Print each line.
+    if os.name=='nt':
+        os.system('cls')
+    else:
         os.system('clear')
     for element in word_list:
         print(element)
@@ -222,6 +229,15 @@ if __name__=="__main__":
     opt = parser.parse_args()
     c = Client(host=opt.host, port=opt.port)
 
+    # while loop to wait for server to be ready
+    print("Waiting for server ready ...")
+    while not START_RECORDING:
+        pass
+    if os.name=='nt':
+        os.system('cls')
+    else:
+        os.system('clear')
+
     if opt.audio is not None:
         resampled_file = resample(opt.audio)
         c.play_file(resampled_file)

requirements/server.txt

@@ -1,5 +1,5 @@
 PyAudio
-faster-whisper
+faster-whisper==0.6.0
 paho-mqtt
 --extra-index-url https://download.pytorch.org/whl/cu113
 torch==1.12.1

server.py

@@ -19,6 +19,7 @@ from transcriber import WhisperModel
 
 
 clients = {}
+SERVER_READY = "SERVER_READY"
 
 def recv_audio(websocket):
     """
@@ -49,7 +50,12 @@ class ServeClient:
         self.payload_size = struct.calcsize("Q")
         self.data = b""
         self.frames = b""
-        self.transcriber = WhisperModel("small.en", compute_type="float16", local_files_only=False)
+        self.transcriber = WhisperModel(
+            "small.en", 
+            device="cuda",
+            compute_type="float16", 
+            local_files_only=False
+        )
         
         # voice activity detection model
         self.vad_model, _ = torch.hub.load(repo_or_dir='snakers4/silero-vad',
@@ -84,6 +90,7 @@ class ServeClient:
         self.websocket = websocket
         self.trans_thread = threading.Thread(target=self.speech_to_text)
         self.trans_thread.start()
+        self.websocket.send(json.dumps(SERVER_READY))
     
     def fill_output(self, output):
         """
@@ -269,7 +276,6 @@ class ServeClient:
         self.transcriber.destroy()
     
 
-
 if __name__ == "__main__":
     with serve(recv_audio, "127.0.0.1", 9090) as server:
         server.serve_forever()