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

[WIP] AWS Nova Sonic service

Browse Source
This commit is contained in:
Paul Kompfner
2025-04-28 09:16:10 -04:00
parent 1f9baefba8
commit 5b64613f65
2 changed files with 14 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
examples/foundational/39-aws-nova-sonic.py
View File

@@ -43,9 +43,7 @@ async def run_bot(webrtc_connection: SmallWebRTCConnection):
camera_in_enabled=False, camera_in_enabled=False,
vad_enabled=True, vad_enabled=True,
vad_audio_passthrough=True, vad_audio_passthrough=True,
# set stop_secs to something roughly similar to the internal setting vad_analyzer=SileroVADAnalyzer(params=VADParams(stop_secs=0.8)),
# of the Multimodal Live api, just to align events.
vad_analyzer=SileroVADAnalyzer(params=VADParams(stop_secs=0.5)),
), ),
) )

22
src/pipecat/services/aws_nova_sonic/aws.py
View File

@@ -137,13 +137,17 @@ class AWSNovaSonicService(LLMService):
await self._send_user_audio_event(frame) await self._send_user_audio_event(frame)
# TODO: do we need to do anything for these? # TODO: do we need to do anything for these?
elif isinstance(frame, StartInterruptionFrame): elif isinstance(frame, StartInterruptionFrame):
print("[pk] StartInterruptionFrame") # print("[pk] StartInterruptionFrame")
pass
elif isinstance(frame, UserStartedSpeakingFrame): elif isinstance(frame, UserStartedSpeakingFrame):
print("[pk] UserStartedSpeakingFrame") # print("[pk] UserStartedSpeakingFrame")
pass
elif isinstance(frame, StopInterruptionFrame): elif isinstance(frame, StopInterruptionFrame):
print("[pk] StopInterruptionFrame") # print("[pk] StopInterruptionFrame")
pass
elif isinstance(frame, UserStoppedSpeakingFrame): elif isinstance(frame, UserStoppedSpeakingFrame):
print("[pk] UserStoppedSpeakingFrame") # print("[pk] UserStoppedSpeakingFrame")
pass
await self.push_frame(frame, direction) await self.push_frame(frame, direction)
@@ -415,7 +419,7 @@ class AWSNovaSonicService(LLMService):
if not self._assistant_is_responding: if not self._assistant_is_responding:
# The assistant has started responding. # The assistant has started responding.
self._assistant_is_responding = True self._assistant_is_responding = True
await self._report_assistant_started_responding() await self._report_assistant_response_started()
async def _handle_text_output_event(self, event_json): async def _handle_text_output_event(self, event_json):
# This should never happen # This should never happen
@@ -471,7 +475,7 @@ class AWSNovaSonicService(LLMService):
# for more information. # for more information.
if not self._assistant_is_responding: if not self._assistant_is_responding:
self._assistant_is_responding = True self._assistant_is_responding = True
await self._report_assistant_started_responding() await self._report_assistant_response_started()
if self._assistant_is_responding: if self._assistant_is_responding:
# Text added to the ongoing assistant response # Text added to the ongoing assistant response
@@ -491,7 +495,7 @@ class AWSNovaSonicService(LLMService):
# last. It's similarly unclear how to determine what the last text content # last. It's similarly unclear how to determine what the last text content
# block will be after an interruption. # block will be after an interruption.
self._assistant_is_responding = False self._assistant_is_responding = False
await self._report_assistant_stopped_responding() await self._report_assistant_response_ended()
elif content.role == Role.USER: elif content.role == Role.USER:
if content.type == ContentType.TEXT: if content.type == ContentType.TEXT:
if content.text_stage == TextStage.FINAL: if content.text_stage == TextStage.FINAL:
@@ -504,7 +508,7 @@ class AWSNovaSonicService(LLMService):
# print("[pk] completion end") # print("[pk] completion end")
pass pass
async def _report_assistant_started_responding(self): async def _report_assistant_response_started(self):
# Report that the assistant has started their response. # Report that the assistant has started their response.
print("[pk] LLM full response started") print("[pk] LLM full response started")
await self.push_frame(LLMFullResponseStartFrame()) await self.push_frame(LLMFullResponseStartFrame())
@@ -522,7 +526,7 @@ class AWSNovaSonicService(LLMService):
print(f"[pk] TTS text: {text}") print(f"[pk] TTS text: {text}")
await self.push_frame(TTSTextFrame(text)) await self.push_frame(TTSTextFrame(text))
async def _report_assistant_stopped_responding(self): async def _report_assistant_response_ended(self):
# Report that the assistant has finished their response. # Report that the assistant has finished their response.
print("[pk] LLM full response ended") print("[pk] LLM full response ended")
await self.push_frame(LLMFullResponseEndFrame()) await self.push_frame(LLMFullResponseEndFrame())