forked from Blendman974/faceclaw
iOS wakeword support
This commit is contained in:
1 parent
d1df4bf046
commit
15a3e129b2
17 files changed
+341
-137
No files matched your search
@@ -130,7 +130,7 @@ public class FaceclawVoiceController {
|
||||
private String lastTranscript = "";
|
||||
private volatile boolean saveRecordings;
|
||||
private volatile boolean endpointing;
|
||||
private final EndpointDetector endpointDetector = new EndpointDetector();
|
||||
private final VoiceEndpointDetector endpointDetector = new VoiceEndpointDetector();
|
||||
private java.io.ByteArrayOutputStream recordingPcm;
|
||||
// Speaker verification against the enrolled wearer voice-print ("my voice
|
||||
// only" command gating). Configured before start(); the utterance PCM is
|
||||
@@ -1028,111 +1028,6 @@ public class FaceclawVoiceController {
|
||||
}
|
||||
}
|
||||
|
||||
/**
|
||||
* Decides when a hands-free utterance is over, so "Hey Even" capture can
|
||||
* stop without a button release.
|
||||
*
|
||||
* Runs on the decoded 16 kHz PCM, so it works the same in every input mode
|
||||
* (the cloud path never sees the samples on this side, and the onboard
|
||||
* recognizer's own endpointing only covers ONBOARD).
|
||||
*
|
||||
* Timing is measured on the sample clock rather than the wall clock: BLE
|
||||
* delivers mic packets in bursts, so elapsed real time badly overestimates
|
||||
* how much audio has actually been heard.
|
||||
*
|
||||
* The threshold is relative to a noise floor measured over the first
|
||||
* {@link #CALIBRATE_MS} of the session, which is roughly the interval where
|
||||
* the user is reacting to the dialog appearing and not yet speaking.
|
||||
*/
|
||||
private static final class EndpointDetector {
|
||||
/** Audio used to estimate the room's noise floor. */
|
||||
private static final int CALIBRATE_MS = 300;
|
||||
/** Speech must exceed this multiple of the noise floor to count as onset. */
|
||||
private static final double ONSET_FACTOR = 3.0;
|
||||
/** Below this multiple of the noise floor counts as silence again. */
|
||||
private static final double RELEASE_FACTOR = 1.8;
|
||||
/** Absolute floor, so a silent room can't make the threshold ~0. */
|
||||
private static final double MIN_RMS = 220.0;
|
||||
/** Trailing silence that ends an utterance. */
|
||||
private static final int SILENCE_MS = 900;
|
||||
/** If the user never speaks, give up rather than record forever. */
|
||||
private static final int LEAD_IN_MS = 6000;
|
||||
/** Hard cap on a single utterance. */
|
||||
private static final int MAX_UTTERANCE_MS = 30000;
|
||||
|
||||
private long totalSamples;
|
||||
private double noiseAccum;
|
||||
private int noisePackets;
|
||||
private double threshold;
|
||||
private boolean speechStarted;
|
||||
private long silenceSamples;
|
||||
private boolean fired;
|
||||
|
||||
void reset() {
|
||||
totalSamples = 0;
|
||||
noiseAccum = 0;
|
||||
noisePackets = 0;
|
||||
threshold = 0;
|
||||
speechStarted = false;
|
||||
silenceSamples = 0;
|
||||
fired = false;
|
||||
}
|
||||
|
||||
/** Returns true exactly once, on the packet that ends the utterance. */
|
||||
boolean accept(short[] pcm, int count) {
|
||||
if (fired || count <= 0) {
|
||||
return false;
|
||||
}
|
||||
totalSamples += count;
|
||||
long elapsedMs = totalSamples * 1000L / SAMPLE_RATE;
|
||||
|
||||
double sumSquares = 0;
|
||||
for (int i = 0; i < count; i++) {
|
||||
double s = pcm[i];
|
||||
sumSquares += s * s;
|
||||
}
|
||||
double rms = Math.sqrt(sumSquares / count);
|
||||
|
||||
if (elapsedMs <= CALIBRATE_MS) {
|
||||
noiseAccum += rms;
|
||||
noisePackets++;
|
||||
return false;
|
||||
}
|
||||
if (threshold == 0) {
|
||||
double noiseFloor = noisePackets > 0 ? noiseAccum / noisePackets : 0;
|
||||
threshold = Math.max(noiseFloor, MIN_RMS);
|
||||
}
|
||||
|
||||
if (!speechStarted) {
|
||||
if (rms >= threshold * ONSET_FACTOR) {
|
||||
speechStarted = true;
|
||||
silenceSamples = 0;
|
||||
} else if (elapsedMs >= LEAD_IN_MS) {
|
||||
// Never heard anything; close the dialog rather than hang.
|
||||
fired = true;
|
||||
return true;
|
||||
}
|
||||
return false;
|
||||
}
|
||||
|
||||
if (rms < threshold * RELEASE_FACTOR) {
|
||||
silenceSamples += count;
|
||||
if (silenceSamples * 1000L / SAMPLE_RATE >= SILENCE_MS) {
|
||||
fired = true;
|
||||
return true;
|
||||
}
|
||||
} else {
|
||||
silenceSamples = 0;
|
||||
}
|
||||
|
||||
if (elapsedMs >= MAX_UTTERANCE_MS) {
|
||||
fired = true;
|
||||
return true;
|
||||
}
|
||||
return false;
|
||||
}
|
||||
}
|
||||
|
||||
private void stopG2Audio() {
|
||||
FaceclawBleCommunicator currentCommunicator = communicator;
|
||||
if (currentCommunicator != null) {
|
||||
|
||||
Reference in new issue
Block a user