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

Smart Turn analyzer now uses the full context of the turn rather than just the audio since VAD last triggered (fixes #3094)

Browse Source
This commit is contained in:
marcus-daily
2025-12-04 11:42:40 +00:00
committed by Marcus
parent b1e5d68d97
commit 7ba0ebba11
2 changed files with 4 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
CHANGELOG.md
View File

@@ -40,6 +40,9 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
- Updated Smart Turn model weights to v3.1. - Updated Smart Turn model weights to v3.1.
- Smart Turn analyzer now uses the full context of the turn rather than just the
audio since VAD last triggered.
### Deprecated ### Deprecated
- Package `pipecat.sync` is deprecated, use `pipecat.utils.sync` instead. - Package `pipecat.sync` is deprecated, use `pipecat.utils.sync` instead.

5
src/pipecat/audio/turn/smart_turn/base_smart_turn.py
View File

@@ -28,7 +28,6 @@ from pipecat.metrics.metrics import MetricsData, SmartTurnMetricsData
STOP_SECS = 3 STOP_SECS = 3
PRE_SPEECH_MS = 0 PRE_SPEECH_MS = 0
MAX_DURATION_SECONDS = 8 # Max allowed segment duration MAX_DURATION_SECONDS = 8 # Max allowed segment duration
USE_ONLY_LAST_VAD_SEGMENT = True
class SmartTurnParams(BaseTurnParams): class SmartTurnParams(BaseTurnParams):
@@ -43,8 +42,6 @@ class SmartTurnParams(BaseTurnParams):
stop_secs: float = STOP_SECS stop_secs: float = STOP_SECS
pre_speech_ms: float = PRE_SPEECH_MS pre_speech_ms: float = PRE_SPEECH_MS
max_duration_secs: float = MAX_DURATION_SECONDS max_duration_secs: float = MAX_DURATION_SECONDS
# not exposing this for now yet until the model can handle it.
# use_only_last_vad_segment: bool = USE_ONLY_LAST_VAD_SEGMENT
class SmartTurnTimeoutException(Exception): class SmartTurnTimeoutException(Exception):
@@ -160,7 +157,7 @@ class BaseSmartTurn(BaseTurnAnalyzer):
state, result = await loop.run_in_executor( state, result = await loop.run_in_executor(
self._executor, self._process_speech_segment, self._audio_buffer self._executor, self._process_speech_segment, self._audio_buffer
) )
if state == EndOfTurnState.COMPLETE or USE_ONLY_LAST_VAD_SEGMENT: if state == EndOfTurnState.COMPLETE:
self._clear(state) self._clear(state)
logger.debug(f"End of Turn result: {state}") logger.debug(f"End of Turn result: {state}")
return state, result return state, result