SAVR: Silence-Aware Voice Recorder

SAVR is a Kotlin library that simplifies audio recording with automatic silence & noise detection and recording termination, using VAD for real-time speech recognision.

Features

• Record audio and save as WAV format to local file recording_${System.currentTimeMillis()}.wav
• Record speech only in real-time (ignore noise & silence) using recordSpeechOnly boolean.
• Once speech is detected, SAVR automatically stops recording after silenceDurationMs of non-speech (noise/silence) input
  • Timer resets every time speech is detected
  • Default value is 5000ms (5 seconds)
• Recording stops entirely after maxRecordingDurationMs
  • Default value is 60000ms (60 seconds)
• Control VAD's speech confidence using vadMode (higher confidence results in fewer false-positives):
  • vadMode = 1 -> 50% confidence
  • vadMode = 2 -> 80% confidence
  • vadMode = 3 -> 95% confidence
• Control minimum duration in milliseconds for silence segments using vadMinimumSilenceDurationMs
  • This parameter defines the necessary and sufficient duration of negative results to recognize it as silence
  • Default value is 300ms (found to work best)
• Control minimum duration in milliseconds for speech segments using vadMinimumSpeechDurationMs
  • This parameter defines the necessary and sufficient duration of positive results to recognize the result as speech
  • Default value is 30ms (found to work best)

Sample App

Run the sample-app to see a demonstration of SAVR in action.

• Simply press "Start Recording", say anything and after silenceDurationMs of silence/noise it will be replayed to you. ("Stop recording" should not be used it is just there for you to manually stop recording if needed).