wx44wx/pipecat
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

refactor party tonight

Browse Source
This commit is contained in:
Moishe Lettvin
2024-01-17 18:42:08 -05:00
parent a3ac0d84e8
commit 13f2f792af
10 changed files with 187 additions and 118 deletions
Download Patch File Download Diff File Expand all files Collapse all files

157
src/dailyai/services/ai_services.py
View File

@@ -2,6 +2,8 @@ import asyncio
import logging
import re
from httpx import request
from dailyai.queue_frame import QueueFrame, FrameType
from abc import abstractmethod
@@ -13,9 +15,7 @@ from collections.abc import Iterable, AsyncIterable
class AIService:
def __init__(
self
):
def __init__(self):
self.logger = logging.getLogger("dailyai")
def stop(self):
@@ -27,30 +27,60 @@ class AIService:
def possible_output_frame_types(self) -> set[FrameType]:
return set()
async def run_to_queue(self, queue: asyncio.Queue, frames, add_end_of_stream=False) -> None:
async for frame in self.run(frames):
print("got frame", frame.frame_type)
await queue.put(frame)
if add_end_of_stream:
await queue.put(QueueFrame(FrameType.END_STREAM, None))
async def run(
self,
requested_frame_types:set[FrameType],
frames:Iterable[QueueFrame] | AsyncIterable[QueueFrame]
) -> AsyncGenerator[QueueFrame, None]:
if self.possible_output_frame_types().intersection(requested_frame_types) == set():
self,
frames: Iterable[QueueFrame]
| AsyncIterable[QueueFrame]
| asyncio.Queue[QueueFrame],
requested_frame_types: set[FrameType] | None=None,
) -> AsyncGenerator[QueueFrame, None]:
if requested_frame_types and self.possible_output_frame_types().intersection(requested_frame_types) == set():
raise Exception(f"Requested frame types {requested_frame_types} are not supported by this service.")
if not requested_frame_types:
requested_frame_types = self.possible_output_frame_types()
print("running", self.__class__.__name__, "with frame types", requested_frame_types)
if isinstance(frames, AsyncIterable):
async for frame in frames:
output_frame: QueueFrame | None = await self.process_frame(requested_frame_types, frame)
if output_frame:
async for output_frame in self.process_frame(requested_frame_types, frame):
print(
"yielding frame", self.__class__.__name__, output_frame.frame_type
)
yield output_frame
elif isinstance(frames, Iterable):
for frame in frames:
output_frame = await self.process_frame(requested_frame_types, frame)
if output_frame:
async for output_frame in self.process_frame(requested_frame_types, frame):
print(
"yielding frame", self.__class__.__name__, output_frame.frame_type
)
yield output_frame
elif isinstance(frames, asyncio.Queue):
while True:
frame = await frames.get()
async for output_frame in self.process_frame(requested_frame_types, frame):
print(
"yielding frame", self.__class__.__name__, output_frame.frame_type
)
yield output_frame
if frame.frame_type == FrameType.END_STREAM:
break
else:
raise Exception("Frames must be an iterable or async iterable")
@abstractmethod
async def process_frame(self, requested_frame_types:set[FrameType], frame:QueueFrame) -> QueueFrame | None:
pass
async def process_frame(self, requested_frame_types:set[FrameType], frame:QueueFrame) -> AsyncGenerator[QueueFrame, None]:
# Yield something so the linter can deduce what should happen here.
yield QueueFrame(FrameType.END_STREAM, None)
class SentenceAggregator(AIService):
def __init__(self, **kwargs):
@@ -63,29 +93,26 @@ class SentenceAggregator(AIService):
def possible_output_frame_types(self) -> set[FrameType]:
return set([FrameType.SENTENCE])
async def process_frame(self, requested_frame_types: set[FrameType], frame: QueueFrame) -> QueueFrame | None:
async def process_frame(self, requested_frame_types: set[FrameType], frame: QueueFrame) -> AsyncGenerator[QueueFrame, None]:
if not FrameType.SENTENCE in requested_frame_types:
return None
return
if frame.frame_type == FrameType.TEXT_CHUNK:
if type(frame.frame_data) != str:
raise Exception("Sentence aggregator requires a string for the data field")
raise Exception(
"Sentence aggregator requires a string for the data field"
)
self.current_sentence += frame.frame_data
if self.current_sentence.endswith((".", "?", "!")):
sentence = self.current_sentence
self.current_sentence = ""
return QueueFrame(FrameType.SENTENCE, sentence)
return None
yield QueueFrame(FrameType.SENTENCE, sentence)
elif frame.frame_type == FrameType.END_STREAM:
if self.current_sentence:
return QueueFrame(FrameType.SENTENCE, self.current_sentence)
else:
return None
yield QueueFrame(FrameType.SENTENCE, self.current_sentence)
elif frame.frame_type == FrameType.SENTENCE:
return frame
else:
return None
yield frame
class LLMService(AIService):
@@ -93,30 +120,29 @@ class LLMService(AIService):
return set([FrameType.LLM_MESSAGE, FrameType.SENTENCE, FrameType.TRANSCRIPTION])
def allowed_output_frame_types(self) -> set[FrameType]:
return set([FrameType.SENTENCE, FrameType.SENTENCE, FrameType.TEXT_CHUNK])
return set([FrameType.SENTENCE, FrameType.TEXT_CHUNK])
async def run_llm_async_sentences(self, messages) -> AsyncGenerator[str, None]:
current_text = ""
async for text in self.run_llm_async(messages):
current_text += text
if re.match(r"^.*[.!?]$", text):
yield current_text
current_text = ""
@abstractmethod
async def run_llm_async(self, messages) -> AsyncGenerator[str, None]:
yield ""
if current_text:
yield current_text
async def process_frame(self, frame:QueueFrame) -> QueueFrame | None:
if not self.output_queue:
raise Exception("Output queue must be set before using the run method.")
@abstractmethod
async def run_llm(self, messages) -> str:
pass
async def process_frame(self, requested_frame_types: set[FrameType], frame: QueueFrame) -> AsyncGenerator[QueueFrame, None]:
if frame.frame_type == FrameType.LLM_MESSAGE:
if type(frame.frame_data) != list:
raise Exception("LLM service requires a dict for the data field")
messages: list[dict[str, str]] = frame.frame_data
async for message in self.run_llm_async_sentences(messages):
await self.output_queue.put(QueueFrame(FrameType.SENTENCE, message))
if FrameType.SENTENCE in requested_frame_types:
yield QueueFrame(FrameType.SENTENCE, await self.run_llm(messages))
else:
async for text_chunk in self.run_llm_async(messages):
yield QueueFrame(FrameType.TEXT_CHUNK, text_chunk)
# TODO: handle other frame types! Need to aggregate into messages
class TTSService(AIService):
@@ -124,6 +150,12 @@ class TTSService(AIService):
def get_mic_sample_rate(self):
return 16000
def allowed_input_frame_types(self) -> set[FrameType]:
return set([FrameType.SENTENCE, FrameType.TRANSCRIPTION, FrameType.TEXT_CHUNK])
def possible_output_frame_types(self) -> set[FrameType]:
return set([FrameType.AUDIO])
# Converts the sentence to audio. Yields a list of audio frames that can
# be sent to the microphone device
@abstractmethod
@@ -131,25 +163,48 @@ class TTSService(AIService):
# yield empty bytes here, so linting can infer what this method does
yield bytes()
async def process_frame(self, frame:QueueFrame) -> QueueFrame | None:
if not self.output_queue:
raise Exception("Output queue must be set before using the run method.")
async def process_frame(self, requested_frame_types: set[FrameType], frame: QueueFrame) -> AsyncGenerator[QueueFrame, None]:
if not FrameType.AUDIO in requested_frame_types:
return
if frame.frame_type == FrameType.SENTENCE:
if type(frame.frame_data) != str:
raise Exception("TTS service requires a string for the data field")
if type(frame.frame_data) != str:
raise Exception("TTS service requires a string for the data field")
text = frame.frame_data
async for audio in self.run_tts(text):
await self.output_queue.put(QueueFrame(FrameType.AUDIO, audio))
async for audio_chunk in self.run_tts(frame.frame_data):
yield QueueFrame(FrameType.AUDIO, audio_chunk)
# Convenience function to send the audio for a sentence to the given queue
async def say(self, sentence, queue: asyncio.Queue):
async for audio_chunk in self.run_tts(sentence):
await queue.put(QueueFrame(FrameType.AUDIO, audio_chunk))
class ImageGenService(AIService):
def __init__(self, image_size, **kwargs):
super().__init__(**kwargs)
self.image_size = image_size
def allowed_input_frame_types(self) -> set[FrameType]:
return set([FrameType.SENTENCE, FrameType.TRANSCRIPTION, FrameType.TEXT_CHUNK, FrameType.IMAGE_DESCRIPTION])
def possible_output_frame_types(self) -> set[FrameType]:
return set([FrameType.IMAGE])
# Renders the image. Returns an Image object.
@abstractmethod
async def run_image_gen(self, sentence, size) -> tuple[str, bytes]:
async def run_image_gen(self, sentence) -> tuple[str, bytes]:
pass
async def process_frame(self, requested_frame_types: set[FrameType], frame: QueueFrame) -> AsyncGenerator[QueueFrame, None]:
if not FrameType.IMAGE in requested_frame_types:
return
if type(frame.frame_data) != str:
raise Exception("Image service requires a string for the data field")
(_, image_data) = await self.run_image_gen(frame.frame_data)
yield QueueFrame(FrameType.IMAGE, image_data)
@dataclass
class AIServiceConfig:

25
src/dailyai/services/azure_ai_services.py
View File

@@ -16,8 +16,8 @@ from PIL import Image
from azure.cognitiveservices.speech import SpeechSynthesizer, SpeechConfig, ResultReason, CancellationReason
class AzureTTSService(TTSService):
def __init__(self, input_queue=None, output_queue=None, speech_key=None, speech_region=None):
super().__init__(input_queue, output_queue)
def __init__(self, speech_key=None, speech_region=None):
super().__init__()
speech_key = speech_key or os.getenv("AZURE_SPEECH_SERVICE_KEY")
speech_region = speech_region or os.getenv("AZURE_SPEECH_SERVICE_REGION")
@@ -35,7 +35,10 @@ class AzureTTSService(TTSService):
"<prosody rate='1.05'>" \
f"{sentence}" \
"</prosody></mstts:express-as></voice></speak> "
result = await asyncio.to_thread(self.speech_synthesizer.speak_ssml, (ssml))
try:
result = await asyncio.to_thread(self.speech_synthesizer.speak_ssml, (ssml))
except Exception as e:
self.logger.error("Error in azure tts", e)
self.logger.info("Got azure tts result")
if result.reason == ResultReason.SynthesizingAudioCompleted:
self.logger.info("Returning result")
@@ -48,8 +51,8 @@ class AzureTTSService(TTSService):
self.logger.info("Error details: {}".format(cancellation_details.error_details))
class AzureLLMService(LLMService):
def __init__(self, input_queue=None, output_queue=None, api_key=None, azure_endpoint=None, api_version=None, model=None):
super().__init__(input_queue, output_queue)
def __init__(self, api_key=None, azure_endpoint=None, api_version=None, model=None):
super().__init__()
api_key = api_key or os.getenv("AZURE_CHATGPT_KEY")
azure_endpoint = azure_endpoint or os.getenv("AZURE_CHATGPT_ENDPOINT")
@@ -92,14 +95,14 @@ class AzureLLMService(LLMService):
class AzureImageGenServiceREST(ImageGenService):
def __init__(self, api_key=None, azure_endpoint=None, api_version=None, model=None):
super().__init__()
def __init__(self, image_size:str, api_key=None, azure_endpoint=None, api_version=None, model=None):
super().__init__(image_size=image_size)
self.api_key = api_key or os.getenv("AZURE_DALLE_KEY")
self.azure_endpoint = azure_endpoint or os.getenv("AZURE_DALLE_ENDPOINT")
self.api_version = api_version or "2023-06-01-preview"
self.model = model or os.getenv("AZURE_DALLE_DEPLOYMENT_ID")
async def run_image_gen(self, sentence, size) -> tuple[str, bytes]:
async def run_image_gen(self, sentence) -> tuple[str, bytes]:
# TODO hoist the session to app-level
async with aiohttp.ClientSession() as session:
url = f"{self.azure_endpoint}openai/images/generations:submit?api-version={self.api_version}"
@@ -107,7 +110,7 @@ class AzureImageGenServiceREST(ImageGenService):
body = {
# Enter your prompt text here
"prompt": sentence,
"size": size,
"size": self.image_size,
"n": 1,
}
async with session.post(url, headers=headers, json=body) as submission:
@@ -153,14 +156,14 @@ class AzureImageGenService(ImageGenService):
api_version=api_version,
)
async def run_image_gen(self, sentence, size) -> tuple[str, bytes]:
async def run_image_gen(self, sentence) -> tuple[str, bytes]:
self.logger.info("Generating azure image", sentence)
image = self.client.images.generate(
model=self.model,
prompt=sentence,
n=1,
size=size,
size=self.image_size,
)
url = image["data"][0]["url"]

3
src/dailyai/services/daily_transport_service.py
View File

@@ -206,6 +206,9 @@ class DailyTransportService(EventHandler):
if frame.frame_type == FrameType.END_STREAM:
break
def wait_for_send_queue_to_empty(self):
self.threadsafe_send_queue.join()
async def run(self) -> None:
self.configure_daily()

4
src/dailyai/services/elevenlabs_ai_service.py
View File

@@ -9,8 +9,8 @@ from dailyai.services.ai_services import TTSService
class ElevenLabsTTSService(TTSService):
def __init__(self, input_queue=None, output_queue=None, api_key=None, voice_id=None):
super().__init__(input_queue, output_queue)
def __init__(self, api_key=None, voice_id=None):
super().__init__()
self.api_key = api_key or os.getenv("ELEVENLABS_API_KEY")
self.voice_id = voice_id or os.getenv("ELEVENLABS_VOICE_ID")

8
src/dailyai/services/open_ai_services.py
View File

@@ -50,20 +50,20 @@ class OpenAILLMService(LLMService):
return None
class OpenAIImageGenService(ImageGenService):
def __init__(self, api_key=None, model=None):
super().__init__()
def __init__(self, image_size:str, api_key=None, model=None):
super().__init__(image_size=image_size)
api_key = api_key or os.getenv("OPEN_AI_KEY")
self.model = model or os.getenv("OPEN_AI_IMAGE_MODEL") or "dall-e-3"
self.client = AsyncOpenAI(api_key=api_key)
async def run_image_gen(self, sentence, size) -> tuple[str, bytes]:
async def run_image_gen(self, sentence) -> tuple[str, bytes]:
self.logger.info("Generating OpenAI image", sentence)
image = await self.client.images.generate(
prompt=sentence,
model=self.model,
n=1,
size=size
size=self.image_size
)
image_url = image.data[0].url
if not image_url: