apunkt/dograh
1
0
Fork 0
mirror of https://github.com/dograh-hq/dograh.git synced 2026-07-22 11:51:04 +02:00
Code Issues Projects Releases Packages Wiki Activity

Feat/enhanced timestamped transcript (#501)

Browse source 
* feat: add additional timestamps in call transcript optionally

* fi: timestamp precision to millisec instead of micro

* fix: address enhanced transcript review issues

* fix: non vad user turn timestamp
This commit is contained in:
Sabiha Khan 2026-07-07 14:47:43 +05:30 • committed by GitHub
parent d9b9a1efc8
commit ac01f7775e
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
12 changed files with 441 additions and 15 deletions
Download patch file Download diff file Expand all files Collapse all files

33
api/services/pipecat/realtime_feedback_observer.py
View file

@ -21,6 +21,7 @@ node changes.
"""
import json
from datetime import UTC, datetime
from typing import TYPE_CHECKING, Awaitable, Callable, Optional, Set
from loguru import logger
@ -52,8 +53,12 @@ from pipecat.frames.frames import (
TranscriptionFrame,
TTSSpeakFrame,
TTSTextFrame,
UserStartedSpeakingFrame,
UserStoppedSpeakingFrame,
UserMuteStartedFrame,
UserMuteStoppedFrame,
VADUserStartedSpeakingFrame,
VADUserStoppedSpeakingFrame,
)
from pipecat.metrics.metrics import TTFBMetricsData
from pipecat.observers.base_observer import BaseObserver, FramePushed
@ -62,6 +67,10 @@ from pipecat.transports.base_output import BaseOutputTransport
from pipecat.utils.enums import RealtimeFeedbackType
def _epoch_seconds_to_utc_iso(timestamp: float) -> str:
return datetime.fromtimestamp(timestamp, UTC).isoformat(timespec="milliseconds")
class RealtimeFeedbackObserver(BaseObserver):
"""Observer that sends real-time events via WebSocket and persists final transcripts.
@ -138,13 +147,35 @@ class RealtimeFeedbackObserver(BaseObserver):
return
# Bot speaking state - WS only (ephemeral state signals, not persisted)
elif isinstance(frame, BotStartedSpeakingFrame):
if self._logs_buffer:
self._logs_buffer.mark_bot_started_speaking()
await self._send_ws(
{"type": RealtimeFeedbackType.BOT_STARTED_SPEAKING.value, "payload": {}}
)
elif isinstance(frame, BotStoppedSpeakingFrame):
if self._logs_buffer:
self._logs_buffer.mark_bot_stopped_speaking()
await self._send_ws(
{"type": RealtimeFeedbackType.BOT_STOPPED_SPEAKING.value, "payload": {}}
)
elif isinstance(frame, UserStartedSpeakingFrame):
if self._logs_buffer:
self._logs_buffer.mark_user_started_speaking()
elif isinstance(frame, UserStoppedSpeakingFrame):
if self._logs_buffer:
self._logs_buffer.mark_user_stopped_speaking()
elif isinstance(frame, VADUserStartedSpeakingFrame):
if self._logs_buffer:
self._logs_buffer.mark_user_started_speaking(
_epoch_seconds_to_utc_iso(frame.timestamp - frame.start_secs),
from_vad=True,
)
elif isinstance(frame, VADUserStoppedSpeakingFrame):
if self._logs_buffer:
self._logs_buffer.mark_user_stopped_speaking(
_epoch_seconds_to_utc_iso(frame.timestamp - frame.stop_secs),
from_vad=True,
)
# User mute state - WS only (ephemeral state signals, not persisted)
elif isinstance(frame, UserMuteStartedFrame):
await self._send_ws(
@ -314,6 +345,7 @@ def register_turn_log_handlers(
text=message.content,
final=True,
timestamp=message.timestamp,
end_timestamp=getattr(message, "end_timestamp", None),
)
)
except Exception as e:
@ -327,6 +359,7 @@ def register_turn_log_handlers(
build_bot_text_event(
text=message.content,
timestamp=message.timestamp,
end_timestamp=getattr(message, "end_timestamp", None),
)
)
except Exception as e: