rename websocket server

server.py

@@ -1,145 +1,82 @@
-import socket, pickle, struct, time, pyaudio
+# import asyncio
+import websockets
+import pickle, struct, time, pyaudio
 import threading
-import os
+import os, json
 import wave
 import textwrap
+
+import logging
+logging.basicConfig(level = logging.INFO)
+
 from collections import deque
 from dataclasses import dataclass
 
-import torch
 import numpy as np
-import paho.mqtt.client as mqtt
+from websockets.sync import server
-
+from websockets.sync.server import serve
 from transcriber import WhisperModel
 
 
-def on_connect(mqttc, obj, flags, rc):
+clients = {}
-    pass
 
-def on_message(mqttc, obj, msg):
+def recv_audio(websocket):
-    pass
+    """
+    Receive audio chunks from client in an infinite loop.
+    """
+    global clients
+    client = ServeClient(websocket)
+    clients[websocket] = client
+    while True:
+        try:
+            frame_data = websocket.recv()
+            if isinstance(frame_data, str):
+                logging.info(frame_data)
+                continue
+            else:
+                frame_np = np.frombuffer(frame_data, np.float32)
+            clients[websocket].add_frames(frame_np)
             
-def on_publish(mqttc, obj, mid):
+        except Exception as e:
-    pass
+            clients[websocket].cleanup()
-
+            clients.pop(websocket)
-def on_subscribe(mqttc, obj, mid, granted_qos):
+            logging.info("Connection Closed.")
-    pass
+            break
-
-def on_log(mqttc, obj, level, string):
-    pass
-
-
-@dataclass(frozen=True)
-class Constants:
-    AUDIO_OVER = b"audio_data_over"
-    ACK = b"acknowledged"
-    SENDING_FILE = b"sending_audio_file"
-    FILE_SENT = b"audio_file_sent"
 
 
 class ServeClient:
-    CHUNK = 1024
-    FORMAT = pyaudio.paInt16
-    CHANNELS = 1
     RATE = 16000
-    def __init__(self, client_socket, device=None, verbose=True):
+    def __init__(self, websocket, topic=None, device=None):
         self.payload_size = struct.calcsize("Q")
         self.data = b""
         self.frames = b""
-        self.frames_np = None
+        self.transcriber = WhisperModel("small.en", compute_type="float16", local_files_only=False)
-        self.transcriber = WhisperModel("medium.en", device="cuda", compute_type="float16")
         self.timestamp_offset = 0.0
+        self.frames_np = None
         self.frames_offset = 0.0
         self.text = []
         self.current_out = ''
         self.prev_out = ''
         self.t_start=None
-        self.client_socket = client_socket
-        self.verbose = verbose
         self.exit = False
         self.same_output_threshold = 0
         self.show_prev_out_thresh = 5   # if pause(no output from whisper) show previous output for 5 seconds
         self.add_pause_thresh = 3       # add a blank to segment list as a pause(no speech) for 3 seconds
+        self.transcript = []
+        self.send_last_n_segments = 10
 
         # text formatting
         self.wrapper = textwrap.TextWrapper(width=50)
         self.pick_previous_segments = 2
 
         # setup mqtt
-        self.topic = None
+        self.topic = topic
-        self.mqttc = mqtt.Client()
-        self.mqttc.on_message = on_message
-        self.mqttc.on_connect = on_connect
-        self.mqttc.on_publish = on_publish
-        self.mqttc.on_subscribe = on_subscribe
-        self.mqttc = mqtt.Client()
-        self.mqttc.connect("mqtt.kurg.org", 1883, 60)
-        self.mqttc.loop_start()
-
-        # send response to client; server is ready
-        self.send_response_to_client(Constants.ACK)
 
         # threading
-        self.recv_thread = threading.Thread(target=self.recv_audio)
+        self.websocket = websocket
         self.trans_thread = threading.Thread(target=self.speech_to_text)
-        self.recv_thread.start()
         self.trans_thread.start()
     
-    def recv_audio(self):
-        """
-        Receive audio chunks from client in an infinite loop.
-        """
-        if self.client_socket:
-            try:
-                while True:
-                    while len(self.data) < self.payload_size:
-                        packet = self.client_socket.recv(4*1024) # 4K
-                        if not packet: break
-                        self.data+=packet
-
-                    packed_msg_size = self.data[:self.payload_size]
-                    self.data = self.data[self.payload_size:]
-                    msg_size = struct.unpack("Q",packed_msg_size)[0]
-                    
-                    while len(self.data) < msg_size:
-                        self.data += self.client_socket.recv(4*1024)
-                    frame_data = self.data[:msg_size]
-                    self.data  = self.data[msg_size:]
-                    frame_data = pickle.loads(frame_data)
-                    if self.topic is None:
-                        self.topic = frame_data["topic"]
-                    
-                    frame = frame_data["audio"]
-                    
-                    # client says audio over
-                    if Constants.AUDIO_OVER in frame:
-                        break
-                    
-                    frame_np = np.frombuffer(frame, dtype=np.int16)
-                    if self.frames_np is not None and self.frames_np.shape[0] > 60*self.RATE:
-                        self.frames_offset += 45.0
-                        self.frames_np = self.frames_np[int(45*self.RATE):]
-                    
-                    if self.frames_np is None:
-                        self.frames_np = frame_np.copy()
-                    else:
-                        self.frames_np = np.concatenate((self.frames_np, frame_np), axis=0)
-
-                # set frames np to None so to stop translation for this client
-                self.frames_np = None
-                self.exit = True
-            except Exception as e:
-                if self.verbose: print(f"[ERROR]: {e}")
-                self.exit = True
-    
-    def send_response_to_client(self, message):
-        """
-        Send serialized response to client.
-        """
-        a = pickle.dumps(message)
-        message = struct.pack("Q",len(a))+a
-        self.client_socket.sendall(message)
-    
     def fill_output(self, output):
         """
         Format output with current and previous complete segments
@@ -159,9 +96,17 @@ class ServeClient:
                 text = ''
             else:
                 text += seg
-        wrapped = self.wrapper.wrap(
+        wrapped = "".join(text + output)
-            text="".join(text + output))[-2:]
+        return wrapped
-        return " ".join(wrapped)
+    
+    def add_frames(self, frame_np):
+        if self.frames_np is not None and self.frames_np.shape[0] > 45*self.RATE:
+            self.frames_offset += 45.0
+            self.frames_np = self.frames_np[int(30*self.RATE):]
+        if self.frames_np is None:
+            self.frames_np = frame_np.copy()
+        else:
+            self.frames_np = np.concatenate((self.frames_np, frame_np), axis=0)
 
     def speech_to_text(self):
         """
@@ -169,30 +114,25 @@ class ServeClient:
         """
         while True:
             if self.exit:
-                self.mqttc.disconnect()
+                logging.info("Exiting speech to text thread")
-                self.client_socket.close()
-                self.transcriber.destroy()
                 break
             
-            if self.frames_np is None: continue
+            if self.frames_np is None: 
+                continue
 
-            # clip audio if the current chunk exceeds 25 seconds, this basically implies that
+            # clip audio if the current chunk exceeds 30 seconds, this basically implies that
-            # no valid segment for the last 25 seconds from whisper
+            # no valid segment for the last 30 seconds from whisper
             if self.frames_np[int((self.timestamp_offset - self.frames_offset)*self.RATE):].shape[0] > 25 * self.RATE:
                 duration = self.frames_np.shape[0] / self.RATE
                 self.timestamp_offset = self.frames_offset + duration - 5
     
-            # add 200 ms from the last chunk if available
+            samples_take = max(0, (self.timestamp_offset - self.frames_offset)*self.RATE)
-            if len(self.text) and self.frames_np[:-int((self.timestamp_offset - self.frames_offset)*self.RATE)].shape[0]:
-                samples_take = max(0, (self.timestamp_offset - self.frames_offset)*self.RATE - 0.2*self.RATE)
-            else:
-                samples_take = max(0, (self.timestamp_offset - self.frames_offset)*self.RATE)
             input_bytes = self.frames_np[int(samples_take):].copy()
             duration = input_bytes.shape[0] / self.RATE
-            if duration<1.0: continue
+            if duration<1.0: 
-
+                continue
             try:
-                input_sample = input_bytes.astype(np.float32) / 32768.0
+                input_sample = input_bytes.copy()
                 # set previous complete segment as initial prompt
                 if len(self.text) and self.text[-1] != '': 
                     initial_prompt = self.text[-1]
@@ -203,37 +143,39 @@ class ServeClient:
                 result = self.transcriber.transcribe(input_sample, initial_prompt=initial_prompt)
                 if len(result):
                     self.t_start = None
-                    output, segments = self.update_segments(result, duration)
+                    last_segment = self.update_segments(result, duration)
-                    out_dict = {
+                    if len(self.transcript) < self.send_last_n_segments:
-                        'text': output,
+                        segments = self.transcript
-                        'segments': segments
+                    else:
-                    }
+                        segments = self.transcript[-self.send_last_n_segments:]
-                    if self.topic is not None:
+                    if last_segment is not None:
-                        self.mqttc.publish(self.topic, payload=str(out_dict))
+                        segments = segments + [last_segment]
-                    self.send_response_to_client(out_dict)
+                    
+                    try:
+                        self.websocket.send(json.dumps(segments))
+                    except Exception as e:
+                        logging.info(f"[ERROR]: {e}")
                 else:
                     # show previous output if there is pause i.e. no output from whisper
-                    output = ''
+                    segments = []
                     if self.t_start is None: self.t_start = time.time()
-
                     if time.time() - self.t_start < self.show_prev_out_thresh:
-                        output = self.fill_output('')
+                        if len(self.transcript) < self.send_last_n_segments:
+                            segments = self.transcript
+                        else:
+                            segments = self.transcript[-self.send_last_n_segments:]
                     
                     # add a blank if there is no speech for 3 seconds
                     if len(self.text) and self.text[-1] != '':
                         if time.time() - self.t_start > self.add_pause_thresh:
                             self.text.append('')
 
-                    # publish outputs
+                    try:
-                    out_dict = {
+                        self.websocket.send(json.dumps(segments))
-                        'text': output,
+                    except Exception as e:
-                        'segments': []
+                        logging.info(f"[INFO]: {e}")
-                    }
-                    if self.topic is not None:
-                        self.mqttc.publish(self.topic, payload=str(out_dict))
-                    self.send_response_to_client(out_dict)
             except Exception as e:
-                if self.verbose: print(f"[ERROR]: {e}")
+                logging.info(f"[INFO]: {e}")
                 time.sleep(0.01)
     
     def update_segments(self, segments, duration):
@@ -249,15 +191,15 @@ class ServeClient:
             transcription for the current chunk
         """
         offset = None
-        transcript = []
         self.current_out = ''
+        last_segment = None
         # process complete segments
         if len(segments) > 1:
             for i, s in enumerate(segments[:-1]):
                 text_ = s.text
                 self.text.append(text_)
                 start, end = self.timestamp_offset + s.start, self.timestamp_offset + min(duration, s.end)
-                transcript.append(
+                self.transcript.append(
                     {
                         'start': start,
                         'end': end,
@@ -268,6 +210,11 @@ class ServeClient:
                 offset = min(duration, s.end)
 
         self.current_out += segments[-1].text
+        last_segment = {
+            'start': self.timestamp_offset + segments[-1].start,
+            'end': self.timestamp_offset + min(duration, segments[-1].end),
+            'text': self.current_out
+        }
         
         # if same incomplete segment is seen multiple times then update the offset
         # and append the segment to the list
@@ -279,7 +226,7 @@ class ServeClient:
         if self.same_output_threshold > 5:
             if not len(self.text) or self.text[-1].strip().lower()!=self.current_out.strip().lower():          
                 self.text.append(self.current_out)
-                transcript.append(
+                self.transcript.append(
                     {
                         'start': self.timestamp_offset,
                         'end': self.timestamp_offset + duration,
@@ -289,6 +236,7 @@ class ServeClient:
             self.current_out = ''
             offset = duration
             self.same_output_threshold = 0
+            last_segment = None
         else:
             self.prev_out = self.current_out
         
@@ -296,36 +244,15 @@ class ServeClient:
         if offset is not None:
             self.timestamp_offset += offset
 
-        # format and return output
+        return last_segment
-        output = self.current_out
+    
-        return self.fill_output(output), transcript
+    def cleanup(self):
+        logging.info("Cleaning up.")
+        self.exit = True
+        self.transcriber.destroy()
 
     
-if __name__=="__main__":
-    # create socket
-    server_socket = socket.socket(socket.AF_INET,socket.SOCK_STREAM)
-    server_socket.setsockopt(socket.SOL_SOCKET, socket.SO_REUSEADDR, 1)
-    host='127.0.0.1'
-    port=5901
-    backlog=5
-    socket_address = (host, port)
-    print('STARTING SERVER AT',socket_address,'...')
-    server_socket.bind(socket_address)
-    server_socket.listen(backlog)
-    client_sockets = []
-    device = 0
-    try:
-        while True:
-            client_socket, addr = server_socket.accept()
-            print('GOT CONNECTION FROM:', addr)
-            client = ServeClient(client_socket, device=f'cuda:{device}')
-            client_sockets.append(client_socket)
-            print("waiting for new connection")
-    except Exception as e:
-        print(f"[ERROR main]: {e}")
-        for sock in client_sockets:
-            try:
-                sock.close()
-            except:
-                pass
 
+if __name__ == "__main__":
+    with serve(recv_audio, "127.0.0.1", 9090) as server:
+        server.serve_forever()

websocket_server.py

@@ -1,258 +0,0 @@
-# import asyncio
-import websockets
-import pickle, struct, time, pyaudio
-import threading
-import os, json
-import wave
-import textwrap
-
-import logging
-logging.basicConfig(level = logging.INFO)
-
-from collections import deque
-from dataclasses import dataclass
-
-import torch
-import numpy as np
-from websockets.sync import server
-from websockets.sync.server import serve
-from transcriber import WhisperModel
-
-
-clients = {}
-
-def recv_audio(websocket):
-    """
-    Receive audio chunks from client in an infinite loop.
-    """
-    global clients
-    client = ServeClient(websocket)
-    clients[websocket] = client
-    while True:
-        try:
-            frame_data = websocket.recv()
-            if isinstance(frame_data, str):
-                logging.info(frame_data)
-                continue
-            frame_np = np.frombuffer(frame_data, np.float32)
-            clients[websocket].add_frames(frame_np)
-            
-        except websockets.ConnectionClosedOK:
-            clients[websocket].cleanup()
-            clients.pop(websocket)
-            logging.info("Connection Closed.")
-            break
-
-
-class ServeClient:
-    RATE = 16000
-    def __init__(self, websocket, topic=None, device=None):
-        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.timestamp_offset = 0.0
-        self.frames_np = None
-        self.frames_offset = 0.0
-        self.text = []
-        self.current_out = ''
-        self.prev_out = ''
-        self.t_start=None
-        self.exit = False
-        self.same_output_threshold = 0
-        self.show_prev_out_thresh = 5   # if pause(no output from whisper) show previous output for 5 seconds
-        self.add_pause_thresh = 3       # add a blank to segment list as a pause(no speech) for 3 seconds
-        self.transcript = []
-        self.send_last_n_segments = 10
-
-        # text formatting
-        self.wrapper = textwrap.TextWrapper(width=50)
-        self.pick_previous_segments = 2
-
-        # setup mqtt
-        self.topic = topic
-
-        # threading
-        self.websocket = websocket
-        self.trans_thread = threading.Thread(target=self.speech_to_text)
-        self.trans_thread.start()
-    
-    def fill_output(self, output):
-        """
-        Format output with current and previous complete segments
-        into two lines of 50 characters.
-
-        Args:
-            output(str): current incomplete segment
-        
-        Returns:
-            transcription wrapped in two lines
-        """
-        text = ''
-        pick_prev = min(len(self.text), self.pick_previous_segments)
-        for seg in self.text[-pick_prev:]:
-            # discard everything before a 3 second pause
-            if seg == '':
-                text = ''
-            else:
-                text += seg
-        wrapped = "".join(text + output)
-        return wrapped
-    
-    def add_frames(self, frame_np):
-        if self.frames_np is not None and self.frames_np.shape[0] > 45*self.RATE:
-            self.frames_offset += 45.0
-            self.frames_np = self.frames_np[int(30*self.RATE):]
-        if self.frames_np is None:
-            self.frames_np = frame_np.copy()
-        else:
-            self.frames_np = np.concatenate((self.frames_np, frame_np), axis=0)
-
-    def speech_to_text(self):
-        """
-        Process audio stream in an infinite loop.
-        """
-        while True:
-            if self.exit:
-                logging.info("Exiting speech to text thread")
-                break
-            
-            if self.frames_np is None: 
-                continue
-
-            # clip audio if the current chunk exceeds 30 seconds, this basically implies that
-            # no valid segment for the last 30 seconds from whisper
-            if self.frames_np[int((self.timestamp_offset - self.frames_offset)*self.RATE):].shape[0] > 25 * self.RATE:
-                duration = self.frames_np.shape[0] / self.RATE
-                self.timestamp_offset = self.frames_offset + duration - 5
-    
-            samples_take = max(0, (self.timestamp_offset - self.frames_offset)*self.RATE)
-            input_bytes = self.frames_np[int(samples_take):].copy()
-            duration = input_bytes.shape[0] / self.RATE
-            if duration<1.0: 
-                continue
-            try:
-                input_sample = input_bytes.copy()
-                # set previous complete segment as initial prompt
-                if len(self.text) and self.text[-1] != '': 
-                    initial_prompt = self.text[-1]
-                else: 
-                    initial_prompt = None
-                
-                # whisper transcribe with prompt
-                result = self.transcriber.transcribe(input_sample, initial_prompt=initial_prompt)
-                if len(result):
-                    self.t_start = None
-                    last_segment = self.update_segments(result, duration)
-                    if len(self.transcript) < self.send_last_n_segments:
-                        segments = self.transcript
-                    else:
-                        segments = self.transcript[-self.send_last_n_segments:]
-                    if last_segment is not None:
-                        segments = segments + [last_segment]
-                    
-                    try:
-                        self.websocket.send(json.dumps(segments))
-                    except Exception as e:
-                        logging.info(f"[ERROR]: {e}")
-                else:
-                    # show previous output if there is pause i.e. no output from whisper
-                    segments = []
-                    if self.t_start is None: self.t_start = time.time()
-                    if time.time() - self.t_start < self.show_prev_out_thresh:
-                        if len(self.transcript) < self.send_last_n_segments:
-                            segments = self.transcript
-                        else:
-                            segments = self.transcript[-self.send_last_n_segments:]
-                    
-                    # add a blank if there is no speech for 3 seconds
-                    if len(self.text) and self.text[-1] != '':
-                        if time.time() - self.t_start > self.add_pause_thresh:
-                            self.text.append('')
-
-                    try:
-                        self.websocket.send(json.dumps(segments))
-                    except Exception as e:
-                        logging.info(f"[INFO]: {e}")
-            except Exception as e:
-                logging.info(f"[INFO]: {e}")
-                time.sleep(0.01)
-    
-    def update_segments(self, segments, duration):
-        """
-        Processes the segments from whisper. Appends all the segments to the list
-        except for the last segment assuming that it is incomplete.
-
-        Args:
-            segments(dict) : dictionary of segments as returned by whisper
-            duration(float): duration of the current chunk
-        
-        Returns:
-            transcription for the current chunk
-        """
-        offset = None
-        self.current_out = ''
-        last_segment = None
-        # process complete segments
-        if len(segments) > 1:
-            for i, s in enumerate(segments[:-1]):
-                text_ = s.text
-                self.text.append(text_)
-                start, end = self.timestamp_offset + s.start, self.timestamp_offset + min(duration, s.end)
-                self.transcript.append(
-                    {
-                        'start': start,
-                        'end': end,
-                        'text': text_
-                    }
-                )
-                
-                offset = min(duration, s.end)
-
-        self.current_out += segments[-1].text
-        last_segment = {
-            'start': self.timestamp_offset + segments[-1].start,
-            'end': self.timestamp_offset + min(duration, segments[-1].end),
-            'text': self.current_out
-        }
-        
-        # if same incomplete segment is seen multiple times then update the offset
-        # and append the segment to the list
-        if self.current_out.strip() == self.prev_out.strip() and self.current_out != '': 
-            self.same_output_threshold += 1
-        else: 
-            self.same_output_threshold = 0
-        
-        if self.same_output_threshold > 5:
-            if not len(self.text) or self.text[-1].strip().lower()!=self.current_out.strip().lower():          
-                self.text.append(self.current_out)
-                self.transcript.append(
-                    {
-                        'start': self.timestamp_offset,
-                        'end': self.timestamp_offset + duration,
-                        'text': self.current_out
-                    }
-                )
-            self.current_out = ''
-            offset = duration
-            self.same_output_threshold = 0
-            last_segment = None
-        else:
-            self.prev_out = self.current_out
-        
-        # update offset
-        if offset is not None:
-            self.timestamp_offset += offset
-
-        return last_segment
-    
-    def cleanup(self):
-        logging.info("Cleaning up.")
-        self.exit = True
-        self.transcriber.destroy()
-
-    
-
-if __name__ == "__main__":
-    with serve(recv_audio, "127.0.0.1", 9090) as server:
-        server.serve_forever()