apunkt/dograh
1
0
Fork 0
mirror of https://github.com/dograh-hq/dograh.git synced 2026-06-07 07:55:16 +02:00
Code Issues Projects Releases Packages Wiki Activity

fix: migrate from custom audio recorder to native AudioBuffer (#115)

Browse source 
* fix: update to pipecat VM Detector

* fix: refactor to remove audio synchronizer

* feat: add speechmatics as STT
This commit is contained in:
Abhishek 2026-01-08 18:03:26 +05:30 • committed by GitHub
parent 31521008cf
commit edf0fa4fbc
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
12 changed files with 193 additions and 591 deletions
Download patch file Download diff file Expand all files Collapse all files

55
api/services/workflow/pipecat_engine.py
View file

@ -1,19 +1,14 @@
from typing import TYPE_CHECKING, Any, Awaitable, Callable, Optional, Union
from api.constants import DEPLOYMENT_MODE, ENABLE_TRACING, VOICEMAIL_RECORDING_DURATION
from api.services.workflow.disposition_mapper import (
apply_disposition_mapping,
get_organization_id_from_workflow_run,
)
from api.services.workflow.pipecat_engine_voicemail_detector import (
VoicemailDetector,
)
from api.services.workflow.workflow import Node, WorkflowGraph
from pipecat.frames.frames import (
CancelFrame,
EndFrame,
FunctionCallResultProperties,
LLMContextFrame,
TTSSpeakFrame,
)
from pipecat.pipeline.task import PipelineTask
@ -93,11 +88,6 @@ class PipecatEngine:
# access to _context
self._variable_extraction_manager = None
# Voicemail detection state
self._detect_voicemail = False
self._voicemail_detector = None
self._voicemail_detection_task: Optional[asyncio.Task] = None
# Lazy loaded built-in function schemas
self._builtin_function_schemas: Optional[list[dict]] = None
@ -172,8 +162,6 @@ class PipecatEngine:
await self.set_node(self.workflow.start_node_id)
# Trigger initial LLM generation
await self.task.queue_frame(LLMContextFrame(self.context))
logger.debug(f"{self.__class__.__name__} initialized")
except Exception as e:
logger.error(f"Error initializing {self.__class__.__name__}: {e}")
@ -388,43 +376,6 @@ class PipecatEngine:
async def _handle_start_node(self, node: Node) -> None:
"""Handle start node execution."""
# Handle voicemail detection setup (before any returns)
# Lets check ENABLE_TRACING to make sure we have prompt access from
# langfuse
if node.detect_voicemail and DEPLOYMENT_MODE == "saas" and ENABLE_TRACING:
if not self._audio_buffer:
logger.warning(
"Voicemail detection enabled but no audio buffer available - skipping detection"
)
else:
logger.debug(
"Start node has detect_voicemail enabled - setting up audio-based detector"
)
self._detect_voicemail = True
self._voicemail_detector = VoicemailDetector(
detection_duration=VOICEMAIL_RECORDING_DURATION,
workflow_run_id=self._workflow_run_id,
)
# Register audio handler on the audio buffer input processor
audio_input = self._audio_buffer.input()
@audio_input.event_handler("on_input_audio_data")
async def handle_voicemail_audio(
processor, pcm, sample_rate, num_channels
):
if (
self._voicemail_detector
and self._voicemail_detector.is_detecting
):
await self._voicemail_detector.handle_audio_data(
processor, pcm, sample_rate, num_channels
)
# Start detection
await self._voicemail_detector.start_detection(self)
# Check if delayed start is enabled
if node.delayed_start:
# Use configured duration or default to 3 seconds
@ -745,8 +696,4 @@ class PipecatEngine:
):
self._user_response_timeout_task.cancel()
# Stop voicemail detection if active
if self._voicemail_detector and hasattr(
self._voicemail_detector, "stop_detection"
):
await self._voicemail_detector.stop_detection()
# Note: Native VoicemailDetector cleanup is handled by the pipeline