io.agora.rtc

Class AgoraAudioVadV2

java.lang.Object

io.agora.rtc.AgoraAudioVadV2

public class AgoraAudioVadV2
extends java.lang.Object

Voice Activity Detection (VAD) V2 implementation for Agora SDK. This is an enhanced version with adaptive RMS threshold support and improved state management.

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
private class	AgoraAudioVadV2.VadData Simple container for an audio frame and its activity flag.

Field Summary

Fields

Modifier and Type	Field and Description
private int	adaptiveStartRmsThreshold Adaptive RMS threshold for start speaking (cached for logging)
private int	adaptiveStopRmsThreshold Adaptive RMS threshold for stop speaking (cached for logging)
private Constants.VadState	currentState Current state of the voice activity detector.
private int	refAvgRmsInLastSession range from 0 to 127, respond to db: -127db, to 0db
private int	silenceCount Count of recent frames considered as silence.
private java.util.Deque<AgoraAudioVadV2.VadData>	startQueue Buffer of frames collected while waiting to start speaking.
private int	startSize Required frame count before declaring the speaking state.
private java.util.Deque<AgoraAudioVadV2.VadData>	stopQueue Buffer of frames collected while confirming the stop state.
private int	totalVoiceRms range from 0 to 127, respond to db: -127db, to 0db
private AgoraAudioVadConfigV2	vadConfig Runtime configuration for the VAD algorithm.
private int	voiceCount Count of recent frames considered as voice.

Constructor Summary

Constructors

Constructor and Description
AgoraAudioVadV2(AgoraAudioVadConfigV2 config) Creates a VAD instance with the specified configuration.

Method Summary

Modifier and Type	Method and Description
void	destroy() Destroys the VAD instance and clears cached data.
int	getAdaptiveStartRmsThreshold() Get current adaptive start RMS threshold
int	getAdaptiveStopRmsThreshold() Get current adaptive stop RMS threshold
private int	getRefActiveAvgRms(AgoraAudioVadConfigV2 currentVadConfig) Get reference average RMS threshold for voice activity detection This implements the adaptive threshold logic from Go SDK
int	getRefAvgRmsInLastSession() Get the reference average RMS from last session
private int	getSilenceCount(java.util.Deque<AgoraAudioVadV2.VadData> queue, int startIndex) Counts the number of silent frames within a queue.
private void	initData() Resets queues and counters according to the current configuration.
boolean	isActive() Check if VAD is currently active (speaking detected)
private boolean	isVadActive(AudioFrame frame, AgoraAudioVadConfigV2 currentVadConfig) Evaluates a frame to decide whether it is active speech.
VadProcessResult	processFrame(AudioFrame frame) Process the audio frame
private byte[]	processSpeaking(AgoraAudioVadV2.VadData vadData, AgoraAudioVadConfigV2 currentVadConfig) Handles frames while speaking and tracks potential stop transitions.
private byte[]	processStart(AgoraAudioVadV2.VadData vadData, AgoraAudioVadConfigV2 currentVadConfig) Handles the transition from silence to speaking by analysing buffered frames.

Methods inherited from class java.lang.Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Field Detail

vadConfig

private volatile AgoraAudioVadConfigV2 vadConfig

Runtime configuration for the VAD algorithm.

currentState

private volatile Constants.VadState currentState

Current state of the voice activity detector.

startSize

private volatile int startSize

Required frame count before declaring the speaking state.

startQueue

private volatile java.util.Deque<AgoraAudioVadV2.VadData> startQueue

Buffer of frames collected while waiting to start speaking.

stopQueue

private volatile java.util.Deque<AgoraAudioVadV2.VadData> stopQueue

Buffer of frames collected while confirming the stop state.

voiceCount

private volatile int voiceCount

Count of recent frames considered as voice.

silenceCount

private volatile int silenceCount

Count of recent frames considered as silence.

totalVoiceRms

private volatile int totalVoiceRms

range from 0 to 127, respond to db: -127db, to 0db

refAvgRmsInLastSession

private volatile int refAvgRmsInLastSession

range from 0 to 127, respond to db: -127db, to 0db

adaptiveStartRmsThreshold

private volatile int adaptiveStartRmsThreshold

Adaptive RMS threshold for start speaking (cached for logging)

adaptiveStopRmsThreshold

private volatile int adaptiveStopRmsThreshold

Adaptive RMS threshold for stop speaking (cached for logging)

Constructor Detail

AgoraAudioVadV2

public AgoraAudioVadV2(AgoraAudioVadConfigV2 config)

Creates a VAD instance with the specified configuration.

Parameters:

config - VAD configuration

Method Detail

initData

private void initData()

Resets queues and counters according to the current configuration.

processFrame

public VadProcessResult processFrame(AudioFrame frame)

Process the audio frame

Parameters:

frame - the audio frame

Returns:

the result of the VAD process

isVadActive

private boolean isVadActive(AudioFrame frame,
                            AgoraAudioVadConfigV2 currentVadConfig)

Evaluates a frame to decide whether it is active speech.

Parameters:

frame - Frame being evaluated

currentVadConfig - Active configuration

Returns:

true if the frame is considered speech

getRefActiveAvgRms

private int getRefActiveAvgRms(AgoraAudioVadConfigV2 currentVadConfig)

Get reference average RMS threshold for voice activity detection This implements the adaptive threshold logic from Go SDK

Parameters:

currentVadConfig - the current VAD configuration

Returns:

the reference RMS threshold value (range: 0-127)

processStart

private byte[] processStart(AgoraAudioVadV2.VadData vadData,
                            AgoraAudioVadConfigV2 currentVadConfig)

Handles the transition from silence to speaking by analysing buffered frames.

Parameters:

vadData - Newly appended data

currentVadConfig - Active configuration

Returns:

Concatenated PCM when a start transition is confirmed; otherwise an empty buffer

processSpeaking

private byte[] processSpeaking(AgoraAudioVadV2.VadData vadData,
                               AgoraAudioVadConfigV2 currentVadConfig)

Handles frames while speaking and tracks potential stop transitions.

Parameters:

vadData - Newly appended data

currentVadConfig - Active configuration

Returns:

Frame data to forward downstream

getSilenceCount

private int getSilenceCount(java.util.Deque<AgoraAudioVadV2.VadData> queue,
                            int startIndex)

Counts the number of silent frames within a queue.

Parameters:

queue - Queue containing historical data

startIndex - Index to begin counting

Returns:

Number of silent frames

isActive

public boolean isActive()

Check if VAD is currently active (speaking detected)

Returns:

true if in speaking state, false otherwise

getRefAvgRmsInLastSession

public int getRefAvgRmsInLastSession()

Get the reference average RMS from last session

Returns:

average RMS value from last speaking session (range: 0-127)

getAdaptiveStartRmsThreshold

public int getAdaptiveStartRmsThreshold()

Get current adaptive start RMS threshold

Returns:

current adaptive start RMS threshold value (range: 0-127)

getAdaptiveStopRmsThreshold

public int getAdaptiveStopRmsThreshold()

Get current adaptive stop RMS threshold

Returns:

current adaptive stop RMS threshold value (range: 0-127)

destroy

public void destroy()

Destroys the VAD instance and clears cached data.