Merge pull request #3 from vocodedev/kian/endpointing

Kian/endpointing

vocode/conversation.py

@@ -16,17 +16,17 @@ from .models.synthesizer import SynthesizerConfig
 from .models.websocket import ReadyMessage, AudioMessage, StartMessage, StopMessage
 from . import api_key
 
-VOCODE_WEBSOCKET_URL = f'wss://api.vocode.dev/conversation'
+VOCODE_WEBSOCKET_URL = f"wss://3fcd-136-24-82-111.ngrok.io/conversation"
+
 
 class Conversation:
-
     def __init__(
         self,
         input_device: BaseInputDevice,
         output_device: BaseOutputDevice,
         transcriber_config: TranscriberConfig,
         agent_config: AgentConfig,
-        synthesizer_config: SynthesizerConfig
+        synthesizer_config: SynthesizerConfig,
     ):
         self.input_device = input_device
         self.output_device = output_device
@@ -55,16 +55,18 @@ class Conversation:
                     await self.output_device.send_async(audio)
                 except queue.Empty:
                     continue
+
         loop = asyncio.new_event_loop()
         loop.run_until_complete(run())
 
     async def start(self):
         async with websockets.connect(f"{VOCODE_WEBSOCKET_URL}?key={api_key}") as ws:
+
             async def sender(ws):
                 start_message = StartMessage(
                     transcriber_config=self.transcriber_config,
                     agent_config=self.agent_config,
-                    synthesizer_config=self.synthesizer_config
+                    synthesizer_config=self.synthesizer_config,
                 )
                 await ws.send(start_message.json())
                 await self.wait_for_ready()
@@ -83,8 +85,6 @@ class Conversation:
                     audio_message = AudioMessage.parse_raw(msg)
                     self.output_audio_queue.put_nowait(audio_message.get_bytes())
 
-
             output_thread = threading.Thread(target=self.play_audio)
             output_thread.start()
             return await asyncio.gather(sender(ws), receiver(ws))
-

vocode/models/transcriber.py

@@ -1,35 +1,51 @@
 from enum import Enum
 from typing import Optional
 from .audio_encoding import AudioEncoding
-from .model import TypedModel
+from .model import BaseModel, TypedModel
 from ..input_device.base_input_device import BaseInputDevice
 
+
 class TranscriberType(str, Enum):
     BASE = "base"
     DEEPGRAM = "deepgram"
     GOOGLE = "google"
     ASSEMBLY_AI = "assembly_ai"
 
+
+class EndpointingConfig(BaseModel):
+    time_cutoff_seconds: Optional[float] = None
+
+
 class TranscriberConfig(TypedModel, type=TranscriberType.BASE):
     sampling_rate: int
     audio_encoding: AudioEncoding
     chunk_size: int
+    endpointing_config: Optional[EndpointingConfig] = None
 
     @classmethod
-    def from_input_device(cls, input_device: BaseInputDevice):
+    def from_input_device(
+        cls,
+        input_device: BaseInputDevice,
+        endpointing_config: Optional[EndpointingConfig] = None,
+    ):
         return cls(
             sampling_rate=input_device.sampling_rate,
             audio_encoding=input_device.audio_encoding,
-            chunk_size=input_device.chunk_size)
+            chunk_size=input_device.chunk_size,
+            endpointing_config=endpointing_config,
+        )
+
 
 class DeepgramTranscriberConfig(TranscriberConfig, type=TranscriberType.DEEPGRAM):
     model: Optional[str] = None
     should_warmup_model: bool = False
     version: Optional[str] = None
 
+
 class GoogleTranscriberConfig(TranscriberConfig, type=TranscriberType.GOOGLE):
     model: Optional[str] = None
     should_warmup_model: bool = False
 
+
 class AssemblyAITranscriberConfig(TranscriberConfig, type=TranscriberType.ASSEMBLY_AI):
     should_warmup_model: bool = False