Closed rosscado closed 1 month ago
Ricky himself writes
Hi Ross Cadogan if you want to make the vad more sensitive you can decrease positiveSpeechThreshold, negativeSpeechThreshold, and minSpeechFrames. If you are passing through the audio to a STT system you can also rely on that to handle false positives
This was improved with an update to the VAD settings in v1.5.14. Say, Pi can now detect single word phrases, while actually getting a bit more accurate on non-speech too.
Users have reported that 'Say, Pi' sometimes does not transmit short phrases, such as single words or very brief utterances, to Pi. This seems to be an issue with the Voice Activity Detection (VAD) system, which is more effective with longer phrases.
Steps to Reproduce:
Expected Behaviour: The extension should be able to detect and transmit short phrases accurately.
Actual Behavior: The extension often requires longer phrases to activate the transmission, leading to difficulty in sending brief utterances.
Additional Information: This issue might be inherent to the sensitivity and configuration of the VAD system, which is calibrated to filter out noise and detect speech. Adjusting its sensitivity or detection thresholds could potentially improve performance with short phrases.
Environment: