first pass at turn based conversation

README.md

@@ -10,9 +10,9 @@ import signal
 
 from vocode.conversation import Conversation
 from vocode.helpers import create_microphone_input_and_speaker_output
-from vocode.models.transcriber import DeepgramTranscriberConfig
-from vocode.models.agent import LLMAgentConfig
-from vocode.models.synthesizer import AzureSynthesizerConfig
+from vocode.streaming.models.transcriber import DeepgramTranscriberConfig
+from vocode.streaming.models.agent import LLMAgentConfig
+from vocode.streaming.models.synthesizer import AzureSynthesizerConfig
 
 if __name__ == "__main__":
     microphone_input, speaker_output = create_microphone_input_and_speaker_output(use_first_available_device=True)

simple_inbound_call_server.py

@@ -1,8 +1,6 @@
-from vocode.telephony.inbound_call_server import InboundCallServer
-from vocode.models.agent import EchoAgentConfig
+from vocode.streaming.telephony.inbound_call_server import InboundCallServer
+from vocode.streaming.models.agent import EchoAgentConfig
 
-if __name__ == '__main__':
-    server = InboundCallServer(
-        agent_config=EchoAgentConfig(initial_message="hello!")
-    )
+if __name__ == "__main__":
+    server = InboundCallServer(agent_config=EchoAgentConfig(initial_message="hello!"))
     server.run(port=3001)

simple_outbound_call.py

@@ -1,14 +1,14 @@
-from vocode.models.synthesizer import AzureSynthesizerConfig
-from vocode.output_device.telephone_output import TelephoneOutput
-from vocode.telephony.outbound_call import OutboundCall
-from vocode.models.telephony import CallEntity
-from vocode.models.agent import (
+from vocode.streaming.models.synthesizer import AzureSynthesizerConfig
+from vocode.streaming.output_device.telephone_output import TelephoneOutput
+from vocode.streaming.telephony.outbound_call import OutboundCall
+from vocode.streaming.models.telephony import CallEntity
+from vocode.streaming.models.agent import (
     EchoAgentConfig,
     ChatGPTAgentConfig,
     WebSocketUserImplementedAgentConfig,
 )
-from vocode.models.message import BaseMessage
-from vocode.telephony.zoom_dial_in import ZoomDialIn
+from vocode.streaming.models.message import BaseMessage
+from vocode.streaming.telephony.zoom_dial_in import ZoomDialIn
 
 if __name__ == "__main__":
     call = ZoomDialIn(
@@ -24,7 +24,7 @@ if __name__ == "__main__":
             generate_responses=True,
             end_conversation_on_goodbye=True,
             send_filler_audio=True,
-            allowed_idle_time_seconds=30
+            allowed_idle_time_seconds=30,
         ),
         synthesizer_config=AzureSynthesizerConfig.from_output_device(
             output_device=TelephoneOutput(), voice_name="en-US-JennyNeural"

simple_streaming_conversation.py

@@ -3,14 +3,14 @@ import logging
 import signal
 from dotenv import load_dotenv
 import os
-from vocode.conversation import Conversation
+from vocode.streaming.streaming_conversation import StreamingConversation
 from vocode.helpers import create_microphone_input_and_speaker_output
-from vocode.models.transcriber import (
+from vocode.streaming.models.transcriber import (
     DeepgramTranscriberConfig,
     PunctuationEndpointingConfig,
     GoogleTranscriberConfig,
 )
-from vocode.models.agent import (
+from vocode.streaming.models.agent import (
     ChatGPTAgentConfig,
     CutOffResponse,
     FillerAudioConfig,
@@ -20,9 +20,9 @@ from vocode.models.agent import (
     LLMAgentConfig,
     ChatGPTAgentConfig,
 )
-from vocode.models.message import BaseMessage
-from vocode.models.synthesizer import AzureSynthesizerConfig
-from vocode.user_implemented_agent.restful_agent import RESTfulAgent
+from vocode.streaming.models.message import BaseMessage
+from vocode.streaming.models.synthesizer import AzureSynthesizerConfig
+from vocode.streaming.user_implemented_agent.restful_agent import RESTfulAgent
 import vocode
 
 load_dotenv()
@@ -34,10 +34,10 @@ logging.root.setLevel(logging.INFO)
 
 if __name__ == "__main__":
     microphone_input, speaker_output = create_microphone_input_and_speaker_output(
-        use_default_devices=False
+        streaming=True, use_default_devices=False
     )
 
-    conversation = Conversation(
+    conversation = StreamingConversation(
         input_device=microphone_input,
         output_device=speaker_output,
         transcriber_config=DeepgramTranscriberConfig.from_input_device(

simple_turn_based_conversation.py

@@ -0,0 +1,33 @@
+import logging
+from dotenv import load_dotenv
+import os
+from vocode.helpers import create_microphone_input_and_speaker_output
+import vocode
+from vocode.turn_based.agent.chat_gpt_agent import ChatGPTAgent
+from vocode.turn_based.synthesizer.azure_synthesizer import AzureSynthesizer
+from vocode.turn_based.transcriber.whisper_transcriber import WhisperTranscriber
+from vocode.turn_based.turn_based_conversation import TurnBasedConversation
+
+load_dotenv()
+vocode.api_key = os.getenv("VOCODE_API_KEY")
+
+
+if __name__ == "__main__":
+    microphone_input, speaker_output = create_microphone_input_and_speaker_output(
+        streaming=False, use_default_devices=False
+    )
+
+    conversation = TurnBasedConversation(
+        input_device=microphone_input,
+        output_device=speaker_output,
+        transcriber=WhisperTranscriber(),
+        agent=ChatGPTAgent(
+            system_prompt="The AI is having a pleasant conversation about life",
+            initial_message="Hello!",
+        ),
+        synthesizer=AzureSynthesizer(sampling_rate=speaker_output.sampling_rate),
+    )
+    while True:
+        conversation.start_speech()
+        input("Press enter to end speech")
+        conversation.end_speech_and_respond()

simple_user_implemented_agent.py

@@ -1,6 +1,6 @@
 from typing import AsyncGenerator
-from vocode.user_implemented_agent.restful_agent import RESTfulAgent
-from vocode.models.agent import (
+from vocode.streaming.user_implemented_agent.restful_agent import RESTfulAgent
+from vocode.streaming.models.agent import (
     RESTfulAgentOutput,
     RESTfulAgentText,
     RESTfulAgentEnd,
@@ -9,7 +9,7 @@ from vocode.models.agent import (
     WebSocketAgentTextMessage,
     WebSocketAgentStopMessage,
 )
-from vocode.user_implemented_agent.websocket_agent import WebSocketAgent
+from vocode.streaming.user_implemented_agent.websocket_agent import WebSocketAgent
 
 
 class TestRESTfulAgent(RESTfulAgent):

vocode/helpers.py

@@ -1,28 +1,69 @@
+from typing import Union
 import sounddevice as sd
-from .input_device.microphone_input import MicrophoneInput
-from .output_device.speaker_output import SpeakerOutput
+from vocode.streaming.input_device.microphone_input import (
+    MicrophoneInput as StreamingMicrophoneInput,
+)
+from vocode.streaming.output_device.speaker_output import (
+    SpeakerOutput as StreamingSpeakerOutput,
+)
+from vocode.turn_based.input_device.microphone_input import (
+    MicrophoneInput as TurnBasedMicrophoneInput,
+)
+from vocode.turn_based.output_device.speaker_output import (
+    SpeakerOutput as TurnBasedSpeakerOutput,
+)
 import logging
 
 logger = logging.getLogger(__name__)
 
+
 def _get_device_prompt(device_infos: list[dict]) -> str:
     return """Please select a device:
 {}
 Choice: """.format(
-        "\n".join(f"{index}: {device['name']}" for index, device in enumerate(device_infos)))
+        "\n".join(
+            f"{index}: {device['name']}" for index, device in enumerate(device_infos)
+        )
+    )
 
-def create_microphone_input_and_speaker_output(use_default_devices=False, mic_sampling_rate=None, speaker_sampling_rate=None) -> tuple[MicrophoneInput, SpeakerOutput]:
+
+def create_microphone_input_and_speaker_output(
+    streaming: bool = True,
+    use_default_devices=False,
+    mic_sampling_rate=None,
+    speaker_sampling_rate=None,
+) -> Union[
+    tuple[StreamingMicrophoneInput, StreamingSpeakerOutput],
+    tuple[TurnBasedMicrophoneInput, TurnBasedSpeakerOutput],
+]:
     device_infos = sd.query_devices()
-    input_device_infos = list(filter(lambda device_info: device_info['max_input_channels'] > 0, device_infos))
-    output_device_infos = list(filter(lambda device_info: device_info['max_output_channels'] > 0, device_infos))
+    input_device_infos = list(
+        filter(lambda device_info: device_info["max_input_channels"] > 0, device_infos)
+    )
+    output_device_infos = list(
+        filter(lambda device_info: device_info["max_output_channels"] > 0, device_infos)
+    )
     if use_default_devices:
-        input_device_info = sd.query_devices(kind='input')
-        output_device_info = sd.query_devices(kind='output')
+        input_device_info = sd.query_devices(kind="input")
+        output_device_info = sd.query_devices(kind="output")
     else:
-        input_device_info = input_device_infos[int(input(_get_device_prompt(input_device_infos)))]
-        output_device_info = output_device_infos[int(input(_get_device_prompt(output_device_infos)))]
-    logger.info("Using microphone input device: %s", input_device_info['name'])
-    microphone_input = MicrophoneInput(input_device_info, sampling_rate=mic_sampling_rate)
-    logger.info("Using speaker output device: %s", output_device_info['name'])
-    speaker_output = SpeakerOutput(output_device_info, sampling_rate=speaker_sampling_rate) 
+        input_device_info = input_device_infos[
+            int(input(_get_device_prompt(input_device_infos)))
+        ]
+        output_device_info = output_device_infos[
+            int(input(_get_device_prompt(output_device_infos)))
+        ]
+    logger.info("Using microphone input device: %s", input_device_info["name"])
+    microphone_class = (
+        StreamingMicrophoneInput if streaming else TurnBasedMicrophoneInput
+    )
+    speaker_class = StreamingSpeakerOutput if streaming else TurnBasedSpeakerOutput
+
+    microphone_input = microphone_class(
+        input_device_info, sampling_rate=mic_sampling_rate
+    )
+    logger.info("Using speaker output device: %s", output_device_info["name"])
+    speaker_output = speaker_class(
+        output_device_info, sampling_rate=speaker_sampling_rate
+    )
     return microphone_input, speaker_output

vocode/streaming/input_device/base_input_device.py

@@ -1,10 +1,12 @@
-from ..models.audio_encoding import AudioEncoding
+from vocode.streaming.models.audio_encoding import AudioEncoding
 import queue
 from typing import Optional
 
-class BaseInputDevice():
 
-    def __init__(self, sampling_rate: int, audio_encoding: AudioEncoding, chunk_size: int):
+class BaseInputDevice:
+    def __init__(
+        self, sampling_rate: int, audio_encoding: AudioEncoding, chunk_size: int
+    ):
         self.sampling_rate = sampling_rate
         self.audio_encoding = audio_encoding
         self.chunk_size = chunk_size

vocode/streaming/input_device/microphone_input.py

@@ -4,25 +4,33 @@ from typing import Optional
 import queue
 import wave
 
-from .base_input_device import BaseInputDevice
-from ..models.audio_encoding import AudioEncoding
+from vocode.streaming.input_device.base_input_device import BaseInputDevice
+from vocode.streaming.models.audio_encoding import AudioEncoding
+
 
 class MicrophoneInput(BaseInputDevice):
-
     DEFAULT_SAMPLING_RATE = 44100
     DEFAULT_CHUNK_SIZE = 2048
 
-    def __init__(self, device_info: dict, sampling_rate: int = None, chunk_size: int = DEFAULT_CHUNK_SIZE, microphone_gain: int = 1):
+    def __init__(
+        self,
+        device_info: dict,
+        sampling_rate: int = None,
+        chunk_size: int = DEFAULT_CHUNK_SIZE,
+        microphone_gain: int = 1,
+    ):
         self.device_info = device_info
-        sampling_rate = sampling_rate or (self.device_info.get('default_samplerate', self.DEFAULT_SAMPLING_RATE))
+        sampling_rate = sampling_rate or (
+            self.device_info.get("default_samplerate", self.DEFAULT_SAMPLING_RATE)
+        )
         super().__init__(int(sampling_rate), AudioEncoding.LINEAR16, chunk_size)
         self.stream = sd.InputStream(
             dtype=np.int16,
             channels=1,
             samplerate=self.sampling_rate,
             blocksize=self.chunk_size,
-            device=int(self.device_info['index']),
-            callback=self._stream_callback
+            device=int(self.device_info["index"]),
+            callback=self._stream_callback,
         )
         self.stream.start()
         self.queue = queue.Queue()

vocode/streaming/input_device/telephone_input.py

@@ -1,5 +1,7 @@
-from vocode.input_device.base_input_device import BaseInputDevice
-from vocode.models.audio_encoding import AudioEncoding
+from vocode.streaming.input_device.base_input_device import (
+    BaseInputDevice,
+)
+from vocode.streaming.models.audio_encoding import AudioEncoding
 
 
 class TelephoneInput(BaseInputDevice):

vocode/streaming/models/agent.py

@@ -3,7 +3,7 @@ from enum import Enum
 
 from pydantic import validator
 
-from vocode.models.message import BaseMessage
+from vocode.streaming.models.message import BaseMessage
 from .model import TypedModel, BaseModel
 
 FILLER_AUDIO_DEFAULT_SILENCE_THRESHOLD_SECONDS = 0.5

vocode/streaming/models/telephony.py

@@ -1,8 +1,8 @@
 from typing import Optional
-from vocode.models.model import BaseModel
-from vocode.models.agent import AgentConfig
-from vocode.models.synthesizer import SynthesizerConfig
-from vocode.models.transcriber import TranscriberConfig
+from vocode.streaming.models.model import BaseModel
+from vocode.streaming.models.agent import AgentConfig
+from vocode.streaming.models.synthesizer import SynthesizerConfig
+from vocode.streaming.models.transcriber import TranscriberConfig
 
 
 class TwilioConfig(BaseModel):

vocode/streaming/models/transcriber.py

@@ -1,8 +1,11 @@
 from enum import Enum
 from typing import Optional
+
+from vocode.streaming.input_device.base_input_device import (
+    BaseInputDevice,
+)
 from .audio_encoding import AudioEncoding
 from .model import BaseModel, TypedModel
-from ..input_device.base_input_device import BaseInputDevice
 
 
 class TranscriberType(str, Enum):

vocode/streaming/output_device/base_output_device.py

@@ -1,7 +1,7 @@
-from ..models.audio_encoding import AudioEncoding
+from vocode.streaming.models.audio_encoding import AudioEncoding
+
 
 class BaseOutputDevice:
-
     def __init__(self, sampling_rate: int, audio_encoding: AudioEncoding):
         self.sampling_rate = sampling_rate
         self.audio_encoding = audio_encoding
@@ -11,5 +11,3 @@ class BaseOutputDevice:
 
     async def maybe_send_mark_async(self, message):
         pass
-
-

vocode/streaming/output_device/speaker_output.py

@@ -2,21 +2,28 @@ import sounddevice as sd
 import numpy as np
 
 from .base_output_device import BaseOutputDevice
-from ..models.audio_encoding import AudioEncoding
+from vocode.streaming.models.audio_encoding import AudioEncoding
+
 
 class SpeakerOutput(BaseOutputDevice):
-
     DEFAULT_SAMPLING_RATE = 44100
 
-    def __init__(self, device_info: dict, sampling_rate: int = None, audio_encoding: AudioEncoding = AudioEncoding.LINEAR16):
+    def __init__(
+        self,
+        device_info: dict,
+        sampling_rate: int = None,
+        audio_encoding: AudioEncoding = AudioEncoding.LINEAR16,
+    ):
         self.device_info = device_info
-        sampling_rate = sampling_rate or int(self.device_info.get('default_samplerate', self.DEFAULT_SAMPLING_RATE))
+        sampling_rate = sampling_rate or int(
+            self.device_info.get("default_samplerate", self.DEFAULT_SAMPLING_RATE)
+        )
         super().__init__(sampling_rate, audio_encoding)
         self.stream = sd.OutputStream(
             channels=1,
             samplerate=self.sampling_rate,
             dtype=np.int16,
-            device=int(self.device_info['index'])
+            device=int(self.device_info["index"]),
         )
         self.stream.start()
 

vocode/streaming/output_device/telephone_output.py

@@ -1,5 +1,5 @@
 from .base_output_device import BaseOutputDevice
-from ..models.audio_encoding import AudioEncoding
+from vocode.streaming.models.audio_encoding import AudioEncoding
 
 
 class TelephoneOutput(BaseOutputDevice):

vocode/streaming/streaming_conversation.py

@@ -8,16 +8,24 @@ import logging
 import threading
 import queue
 import vocode
-from vocode.input_device.base_input_device import BaseInputDevice
-from vocode.output_device.base_output_device import BaseOutputDevice
-from vocode.models.transcriber import TranscriberConfig
-from vocode.models.agent import AgentConfig
-from vocode.models.synthesizer import SynthesizerConfig
-from vocode.models.websocket import ReadyMessage, AudioMessage, StartMessage, StopMessage
+from vocode.streaming.input_device.base_input_device import (
+    BaseInputDevice,
+)
+from vocode.streaming.output_device.base_output_device import BaseOutputDevice
+from vocode.streaming.models.transcriber import TranscriberConfig
+from vocode.streaming.models.agent import AgentConfig
+from vocode.streaming.models.synthesizer import SynthesizerConfig
+from vocode.streaming.models.websocket import (
+    ReadyMessage,
+    AudioMessage,
+    StartMessage,
+    StopMessage,
+)
 
 load_dotenv()
 
-class Conversation:
+
+class StreamingConversation:
     def __init__(
         self,
         input_device: BaseInputDevice,
@@ -61,14 +69,16 @@ class Conversation:
         loop.run_until_complete(run())
 
     async def start(self):
-        async with websockets.connect(f"{self.vocode_websocket_url}?key={vocode.api_key}") as ws:
+        async with websockets.connect(
+            f"{self.vocode_websocket_url}?key={vocode.api_key}"
+        ) as ws:
 
             async def sender(ws: WebSocketClientProtocol):
                 start_message = StartMessage(
                     transcriber_config=self.transcriber_config,
                     agent_config=self.agent_config,
                     synthesizer_config=self.synthesizer_config,
-                    conversation_id=self.id
+                    conversation_id=self.id,
                 )
                 await ws.send(start_message.json())
                 await self.wait_for_ready()

vocode/streaming/telephony/inbound_call_server.py

@@ -4,10 +4,15 @@ import requests
 import uvicorn
 
 import vocode
-from vocode.models.transcriber import TranscriberConfig
-from vocode.models.synthesizer import SynthesizerConfig
-from vocode.models.agent import AgentConfig
-from vocode.models.telephony import CreateInboundCall, TwilioConfig, TwilioConfig
+from vocode.streaming.models.transcriber import TranscriberConfig
+from vocode.streaming.models.synthesizer import SynthesizerConfig
+from vocode.streaming.models.agent import AgentConfig
+from vocode.streaming.models.telephony import (
+    CreateInboundCall,
+    TwilioConfig,
+    TwilioConfig,
+)
+
 
 class InboundCallServer:
     def __init__(

vocode/streaming/telephony/outbound_call.py

@@ -2,9 +2,9 @@ from typing import Optional
 import requests
 
 import vocode
-from vocode.models.agent import AgentConfig
-from vocode.models.synthesizer import SynthesizerConfig
-from vocode.models.transcriber import TranscriberConfig
+from vocode.streaming.models.agent import AgentConfig
+from vocode.streaming.models.synthesizer import SynthesizerConfig
+from vocode.streaming.models.transcriber import TranscriberConfig
 from ..models.telephony import (
     CallEntity,
     CreateOutboundCall,
@@ -31,8 +31,12 @@ class OutboundCall:
         self.synthesizer_config = synthesizer_config
         self.conversation_id = conversation_id
         self.twilio_config = twilio_config
-        self.vocode_create_outbound_call_url = f"https://{vocode.base_url}/create_outbound_call"
-        self.vocode_end_outbound_call_url = f"https://{vocode.base_url}/end_outbound_call"
+        self.vocode_create_outbound_call_url = (
+            f"https://{vocode.base_url}/create_outbound_call"
+        )
+        self.vocode_end_outbound_call_url = (
+            f"https://{vocode.base_url}/end_outbound_call"
+        )
 
     def start(self) -> str:
         response = requests.post(

vocode/streaming/telephony/zoom_dial_in.py

@@ -2,11 +2,11 @@ from typing import Optional
 import requests
 
 import vocode
-from vocode.models.agent import AgentConfig
-from vocode.models.synthesizer import SynthesizerConfig
-from vocode.models.transcriber import TranscriberConfig
-from vocode.telephony.outbound_call import OutboundCall
-from vocode.models.telephony import (
+from vocode.streaming.models.agent import AgentConfig
+from vocode.streaming.models.synthesizer import SynthesizerConfig
+from vocode.streaming.models.transcriber import TranscriberConfig
+from vocode.streaming.telephony.outbound_call import OutboundCall
+from vocode.streaming.models.telephony import (
     CallEntity,
     DialIntoZoomCall,
     TwilioConfig,

vocode/turn_based/agent/base_agent.py

@@ -0,0 +1,9 @@
+from typing import Optional
+
+
+class BaseAgent:
+    def __init__(self, initial_message: Optional[str] = None):
+        self.initial_message = initial_message
+
+    def respond(self, human_input: str):
+        raise NotImplementedError

vocode/turn_based/agent/chat_gpt_agent.py

@@ -0,0 +1,45 @@
+from typing import Optional
+from langchain.prompts import (
+    ChatPromptTemplate,
+    MessagesPlaceholder,
+    SystemMessagePromptTemplate,
+    HumanMessagePromptTemplate,
+)
+from langchain.chains import ConversationChain
+from langchain.chat_models import ChatOpenAI
+from langchain.memory import ConversationBufferMemory
+
+from vocode.turn_based.agent.base_agent import BaseAgent
+
+
+class ChatGPTAgent(BaseAgent):
+    def __init__(
+        self,
+        system_prompt: str,
+        initial_message: Optional[str] = None,
+        model_name: str = "gpt-3.5-turbo",
+        temperature: float = 0.7,
+        max_tokens: int = 100,
+    ):
+        super().__init__(initial_message=initial_message)
+        self.prompt = ChatPromptTemplate.from_messages(
+            [
+                SystemMessagePromptTemplate.from_template(system_prompt),
+                MessagesPlaceholder(variable_name="history"),
+                HumanMessagePromptTemplate.from_template("{input}"),
+            ]
+        )
+        self.memory = ConversationBufferMemory(return_messages=True)
+        if initial_message:
+            self.memory.chat_memory.add_ai_message(initial_message)
+        self.llm = ChatOpenAI(
+            model_name=model_name,
+            temperature=temperature,
+            max_tokens=max_tokens,
+        )
+        self.conversation = ConversationChain(
+            memory=self.memory, prompt=self.prompt, llm=self.llm
+        )
+
+    def respond(self, human_input: str):
+        return self.conversation.predict(input=human_input)

vocode/turn_based/agent/echo_agent.py

@@ -0,0 +1,6 @@
+from vocode.turn_based.agent.base_agent import BaseAgent
+
+
+class EchoAgent(BaseAgent):
+    def respond(self, human_input: str):
+        return human_input

vocode/turn_based/input_device/base_input_device.py

@@ -0,0 +1,9 @@
+from pydub import AudioSegment
+
+
+class BaseInputDevice:
+    def start_listening(self):
+        raise NotImplementedError
+
+    def end_listening(self) -> AudioSegment:
+        raise NotImplementedError

vocode/turn_based/input_device/microphone_input.py

@@ -0,0 +1,59 @@
+from typing import Optional
+import sounddevice as sd
+import numpy as np
+from pydub import AudioSegment
+import io
+import wave
+
+from vocode.turn_based.input_device.base_input_device import BaseInputDevice
+
+
+class MicrophoneInput(BaseInputDevice):
+    DEFAULT_SAMPLING_RATE = 44100
+    DEFAULT_CHUNK_SIZE = 2048
+
+    def __init__(
+        self,
+        device_info: dict,
+        sampling_rate: int = None,
+        chunk_size: int = DEFAULT_CHUNK_SIZE,
+    ):
+        self.device_info = device_info
+        self.sampling_rate = sampling_rate or (
+            self.device_info.get("default_samplerate", self.DEFAULT_SAMPLING_RATE)
+        )
+        self.chunk_size = chunk_size
+        self.buffer: Optional[io.BytesIO] = None
+        self.wave_writer: Optional[wave.Wave_write] = None
+
+    def create_stream(self):
+        return sd.InputStream(
+            dtype=np.int16,
+            channels=1,
+            samplerate=self.sampling_rate,
+            blocksize=self.chunk_size,
+            device=int(self.device_info["index"]),
+            callback=self._stream_callback,
+        )
+
+    def _stream_callback(self, in_data: np.ndarray[np.int16], *_args):
+        audio_bytes = in_data.tobytes()
+        self.wave_writer.writeframes(audio_bytes)
+
+    def create_buffer(self):
+        in_memory_wav = io.BytesIO()
+        wave_writer = wave.open(in_memory_wav, "wb")
+        wave_writer.setnchannels(1)
+        wave_writer.setsampwidth(2)
+        wave_writer.setframerate(self.sampling_rate)
+        return in_memory_wav, wave_writer
+
+    def start_listening(self):
+        self.buffer, self.wave_writer = self.create_buffer()
+        self.stream = self.create_stream()
+        self.stream.start()
+
+    def end_listening(self) -> AudioSegment:
+        self.stream.stop()
+        self.buffer.seek(0)
+        return AudioSegment.from_wav(self.buffer)

vocode/turn_based/output_device/base_output_device.py

@@ -0,0 +1,9 @@
+from pydub import AudioSegment
+
+
+class BaseOutputDevice:
+    def send_audio(self, audio: AudioSegment) -> None:
+        raise NotImplementedError
+
+    def terminate(self):
+        pass

vocode/turn_based/output_device/speaker_output.py

@@ -0,0 +1,32 @@
+import sounddevice as sd
+import numpy as np
+from pydub import AudioSegment
+
+from vocode.turn_based.output_device.base_output_device import BaseOutputDevice
+
+
+class SpeakerOutput(BaseOutputDevice):
+    DEFAULT_SAMPLING_RATE = 44100
+
+    def __init__(
+        self,
+        device_info: dict,
+        sampling_rate: int = None,
+    ):
+        self.device_info = device_info
+        self.sampling_rate = sampling_rate or int(
+            self.device_info.get("default_samplerate", self.DEFAULT_SAMPLING_RATE)
+        )
+        self.stream = sd.OutputStream(
+            channels=1,
+            samplerate=self.sampling_rate,
+            dtype=np.int16,
+            device=int(self.device_info["index"]),
+        )
+        self.stream.start()
+
+    def send_audio(self, audio_segment: AudioSegment):
+        self.stream.write(np.frombuffer(audio_segment.raw_data, dtype=np.int16))
+
+    def terminate(self):
+        self.stream.close()

vocode/turn_based/synthesizer/azure_synthesizer.py

@@ -0,0 +1,53 @@
+import os
+from dotenv import load_dotenv
+import azure.cognitiveservices.speech as speechsdk
+from pydub import AudioSegment
+
+from vocode.turn_based.synthesizer.base_synthesizer import BaseSynthesizer
+
+load_dotenv()
+
+
+class AzureSynthesizer(BaseSynthesizer):
+    def __init__(self, sampling_rate: int):
+        self.sampling_rate = sampling_rate
+        speech_config = speechsdk.SpeechConfig(
+            subscription=os.environ.get("AZURE_SPEECH_KEY"),
+            region=os.environ.get("AZURE_SPEECH_REGION"),
+        )
+        if self.sampling_rate == 44100:
+            speech_config.set_speech_synthesis_output_format(
+                speechsdk.SpeechSynthesisOutputFormat.Raw44100Hz16BitMonoPcm
+            )
+        if self.sampling_rate == 48000:
+            speech_config.set_speech_synthesis_output_format(
+                speechsdk.SpeechSynthesisOutputFormat.Raw48Khz16BitMonoPcm
+            )
+        if self.sampling_rate == 24000:
+            speech_config.set_speech_synthesis_output_format(
+                speechsdk.SpeechSynthesisOutputFormat.Raw24Khz16BitMonoPcm
+            )
+        elif self.sampling_rate == 16000:
+            speech_config.set_speech_synthesis_output_format(
+                speechsdk.SpeechSynthesisOutputFormat.Raw16Khz16BitMonoPcm
+            )
+        elif self.sampling_rate == 8000:
+            speech_config.set_speech_synthesis_output_format(
+                speechsdk.SpeechSynthesisOutputFormat.Raw8Khz16BitMonoPcm
+            )
+
+        self.synthesizer = speechsdk.SpeechSynthesizer(
+            speech_config=speech_config, audio_config=None
+        )
+
+    def synthesize(self, text) -> AudioSegment:
+        result = self.synthesizer.speak_text(text)
+        if result.reason == speechsdk.ResultReason.SynthesizingAudioCompleted:
+            return AudioSegment(
+                result.audio_data,
+                sample_width=2,
+                frame_rate=self.sampling_rate,
+                channels=1,
+            )
+        else:
+            raise Exception("Could not synthesize audio")

vocode/turn_based/synthesizer/base_synthesizer.py

@@ -0,0 +1,6 @@
+from pydub import AudioSegment
+
+
+class BaseSynthesizer:
+    def synthesize(self, text) -> AudioSegment:
+        raise NotImplementedError

vocode/turn_based/transcriber/base_transcriber.py

@@ -0,0 +1,6 @@
+from pydub import AudioSegment
+
+
+class BaseTranscriber:
+    def transcribe(self, audio_segment: AudioSegment) -> str:
+        raise NotImplementedError

vocode/turn_based/transcriber/whisper_transcriber.py

@@ -0,0 +1,21 @@
+from pydub import AudioSegment
+import io
+import os
+from dotenv import load_dotenv
+import openai
+
+from vocode.turn_based.transcriber.base_transcriber import BaseTranscriber
+
+load_dotenv()
+
+openai.api_key = os.getenv("OPENAI_API_KEY")
+
+
+class WhisperTranscriber(BaseTranscriber):
+    def transcribe(self, audio_segment: AudioSegment) -> str:
+        in_memory_wav = io.BytesIO()
+        audio_segment.export(in_memory_wav, format="wav")
+        in_memory_wav.seek(0)
+        in_memory_wav.name = "whisper.wav"
+        transcript = openai.Audio.transcribe("whisper-1", in_memory_wav)
+        return transcript.text

vocode/turn_based/turn_based_conversation.py

@@ -0,0 +1,38 @@
+from vocode.turn_based.agent.base_agent import BaseAgent
+from vocode.turn_based.input_device.base_input_device import (
+    BaseInputDevice,
+)
+from vocode.turn_based.output_device.base_output_device import BaseOutputDevice
+from vocode.turn_based.synthesizer.base_synthesizer import BaseSynthesizer
+from vocode.turn_based.transcriber.base_transcriber import BaseTranscriber
+
+
+class TurnBasedConversation:
+    def __init__(
+        self,
+        input_device: BaseInputDevice,
+        transcriber: BaseTranscriber,
+        agent: BaseAgent,
+        synthesizer: BaseSynthesizer,
+        output_device: BaseOutputDevice,
+    ):
+        self.input_device = input_device
+        self.transcriber = transcriber
+        self.agent = agent
+        self.synthesizer = synthesizer
+        self.output_device = output_device
+        self.maybe_play_initial_message()
+
+    def maybe_play_initial_message(self):
+        if self.agent.initial_message:
+            self.output_device.send_audio(
+                self.synthesizer.synthesize(self.agent.initial_message)
+            )
+
+    def start_speech(self):
+        self.input_device.start_listening()
+
+    def end_speech_and_respond(self):
+        human_input = self.transcriber.transcribe(self.input_device.end_listening())
+        agent_response = self.agent.respond(human_input)
+        self.output_device.send_audio(self.synthesizer.synthesize(agent_response))