From d14fa75079de570cb9c4dbf63f6b5bd65f959b09 Mon Sep 17 00:00:00 2001
From: Ariver <shanghai3168@gmail.com>
Date: Sun, 14 Jun 2026 20:14:34 +0800
Subject: [PATCH] Fix X-ASR release tail capture
---
privatevoice.src/internal/engine/engine_darwin.go | 6 ++
privatevoice.src/internal/engine/engine_darwin_test.go | 14 ++++
privatevoice.src/app.go | 95 +++++++++++++++++++++++--------
privatevoice.src/app_live_caption_test.go | 54 ++++++++++++++++++
privatevoice.src/internal/engine/engine.go | 7 ++
5 files changed, 151 insertions(+), 25 deletions(-)
diff --git a/privatevoice.src/app.go b/privatevoice.src/app.go
index 73a8aec..7412d0a 100755
--- a/privatevoice.src/app.go
+++ b/privatevoice.src/app.go
@@ -63,23 +63,25 @@
indicator overlay.Overlay
// State
- mu sync.Mutex
- isRecording bool
- isFreetalking bool
- hotkeyActive bool
- hotkeyPressTime time.Time
- isCombination bool
- tapStopOnPress bool
- isRecordingHotkey bool
- hideGen atomic.Uint64
- lastStopTime time.Time
- hotkeyPollSeen bool
- lastHotkeyDown bool
- lastHotkeyBlocked time.Time
- silenceSince time.Time // when continuous silence started (free-talk only)
- freeTalkStart time.Time // when free-talk mode started
- liveCaptionCancel context.CancelFunc
- liveCaptionSeq uint64
+ mu sync.Mutex
+ isRecording bool
+ isStoppingRecording bool
+ isFreetalking bool
+ hotkeyActive bool
+ hotkeyPressTime time.Time
+ isCombination bool
+ tapStopOnPress bool
+ isRecordingHotkey bool
+ hideGen atomic.Uint64
+ lastStopTime time.Time
+ hotkeyPollSeen bool
+ lastHotkeyDown bool
+ lastHotkeyBlocked time.Time
+ silenceSince time.Time // when continuous silence started (free-talk only)
+ freeTalkStart time.Time // when free-talk mode started
+ liveCaptionCancel context.CancelFunc
+ liveCaptionSeq uint64
+ releaseTailCaptureNanos atomic.Int64
}
func RunApp() error {
@@ -279,10 +281,10 @@
)
}
- if a.eng == nil || a.isRecordingHotkey {
+ if a.eng == nil || a.isRecordingHotkey || a.isStoppingRecording {
if isDown && time.Since(a.lastHotkeyBlocked) > time.Second {
a.lastHotkeyBlocked = time.Now()
- logger.Info("Hotkey ignored: key=%s engineReady=%t recordingHotkey=%t", hotkey.GetKeyName(a.cfg.HotkeyVK), a.eng != nil, a.isRecordingHotkey)
+ logger.Info("Hotkey ignored: key=%s engineReady=%t recordingHotkey=%t stoppingRecording=%t", hotkey.GetKeyName(a.cfg.HotkeyVK), a.eng != nil, a.isRecordingHotkey, a.isStoppingRecording)
}
return
}
@@ -407,7 +409,7 @@
}
func (a *App) startRecordingLocked() {
- if a.isRecording {
+ if a.isRecording || a.isStoppingRecording {
return
}
a.isRecording = true
@@ -450,16 +452,16 @@
}
a.stopLiveCaptionLocked()
- hasVoice := a.recorder.HasVoiceActivity()
+ a.isStoppingRecording = true
a.indicator.SetStatus(overlay.StatusProcessing, "识别中")
go func() {
- samples := a.recorder.StopAndGetSamples()
+ samples, hasVoice := a.stopRecorderAfterReleaseTailCapture()
a.recognizeAndPaste(hasVoice, samples)
}()
}
func (a *App) startTapRecordingLocked() {
- if a.isRecording || a.isFreetalking {
+ if a.isRecording || a.isStoppingRecording || a.isFreetalking {
return
}
a.isFreetalking = true
@@ -625,11 +627,11 @@
a.lastStopTime = time.Now()
a.stopLiveCaptionLocked()
- hasVoice := a.recorder.HasVoiceActivity()
+ a.isStoppingRecording = true
logger.Info("Tap recording stopped")
a.indicator.SetStatus(overlay.StatusProcessing, "识别中")
go func() {
- samples := a.recorder.StopAndGetSamples()
+ samples, hasVoice := a.stopRecorderAfterReleaseTailCapture()
a.recognizeAndPaste(hasVoice, samples)
}()
}
@@ -688,6 +690,35 @@
logger.Info("Free talk stopped (silence auto-stop)")
a.indicator.SetStatus(overlay.StatusProcessing, "识别中")
a.recognizeAndPaste(hasVoice, samples)
+}
+
+func (a *App) waitForReleaseTailCapture() {
+ delay := a.releaseTailCaptureDelay()
+ if delay <= 0 {
+ return
+ }
+ logger.Info("Release tail capture: waiting %dms before stopping audio", delay.Milliseconds())
+ time.Sleep(delay)
+}
+
+func (a *App) releaseTailCaptureDelay() time.Duration {
+ nanos := a.releaseTailCaptureNanos.Load()
+ if nanos <= 0 {
+ return 0
+ }
+ return time.Duration(nanos)
+}
+
+func (a *App) stopRecorderAfterReleaseTailCapture() ([]float32, bool) {
+ a.waitForReleaseTailCapture()
+ samples := a.recorder.StopAndGetSamples()
+ hasVoice := a.recorder.HasVoiceActivity()
+
+ a.mu.Lock()
+ a.isStoppingRecording = false
+ a.mu.Unlock()
+
+ return samples, hasVoice
}
// recognizeAndPaste runs ASR on the recorded samples and pastes the result.
@@ -857,6 +888,8 @@
}
func (a *App) replaceEngine(eng engine.Engine) {
+ a.releaseTailCaptureNanos.Store(int64(releaseTailCaptureDelayForEngine(eng)))
+
a.engineMu.Lock()
old := a.eng
a.eng = eng
@@ -867,6 +900,18 @@
}
}
+func releaseTailCaptureDelayForEngine(eng engine.Engine) time.Duration {
+ tailCaptureEng, ok := eng.(engine.ReleaseTailCaptureEngine)
+ if !ok {
+ return 0
+ }
+ delay := tailCaptureEng.ReleaseTailCaptureDelay()
+ if delay < 0 {
+ return 0
+ }
+ return delay
+}
+
func hotkeyReadyText(keyName, mode string) string {
if mode == config.HotkeyModeTap {
return "点按" + keyName + "说话"
diff --git a/privatevoice.src/app_live_caption_test.go b/privatevoice.src/app_live_caption_test.go
index 55b8316..b8804c1 100644
--- a/privatevoice.src/app_live_caption_test.go
+++ b/privatevoice.src/app_live_caption_test.go
@@ -3,6 +3,7 @@
import (
"strings"
"testing"
+ "time"
)
func TestLiveCaptionDisplayTextFitsOverlayBuffer(t *testing.T) {
@@ -18,3 +19,56 @@
t.Fatalf("caption = %q, want leading truncation marker", got)
}
}
+
+func TestReleaseTailCaptureDelayUsesEngineCapability(t *testing.T) {
+ a := &App{}
+ a.replaceEngine(tailCaptureTestEngine{delay: 300 * time.Millisecond})
+
+ if got := a.releaseTailCaptureDelay(); got != 300*time.Millisecond {
+ t.Fatalf("releaseTailCaptureDelay() = %v, want 300ms", got)
+ }
+}
+
+func TestReleaseTailCaptureDelayDefaultsToZero(t *testing.T) {
+ a := &App{}
+ a.replaceEngine(plainTestEngine{})
+
+ if got := a.releaseTailCaptureDelay(); got != 0 {
+ t.Fatalf("releaseTailCaptureDelay() = %v, want 0", got)
+ }
+}
+
+func TestStartRecordingIgnoredWhileStoppingRecording(t *testing.T) {
+ a := &App{isStoppingRecording: true}
+
+ a.startRecordingLocked()
+
+ if a.isRecording {
+ t.Fatal("startRecordingLocked should not start while stop tail capture is pending")
+ }
+}
+
+func TestStartTapRecordingIgnoredWhileStoppingRecording(t *testing.T) {
+ a := &App{isStoppingRecording: true}
+
+ a.startTapRecordingLocked()
+
+ if a.isRecording || a.isFreetalking {
+ t.Fatal("startTapRecordingLocked should not start while stop tail capture is pending")
+ }
+}
+
+type plainTestEngine struct{}
+
+func (plainTestEngine) Recognize([]float32) (string, error) { return "", nil }
+func (plainTestEngine) HardwareInfo() string { return "test" }
+func (plainTestEngine) Close() {}
+
+type tailCaptureTestEngine struct {
+ plainTestEngine
+ delay time.Duration
+}
+
+func (e tailCaptureTestEngine) ReleaseTailCaptureDelay() time.Duration {
+ return e.delay
+}
diff --git a/privatevoice.src/internal/engine/engine.go b/privatevoice.src/internal/engine/engine.go
index 3f939c1..c3e564d 100755
--- a/privatevoice.src/internal/engine/engine.go
+++ b/privatevoice.src/internal/engine/engine.go
@@ -2,6 +2,7 @@
import (
"fmt"
+ "time"
"voicesnap/internal/config"
"voicesnap/internal/language"
"voicesnap/internal/logger"
@@ -26,6 +27,12 @@
NewStreamingSession() (StreamingSession, error)
}
+// ReleaseTailCaptureEngine is implemented by engines that need a short audio
+// capture grace period after the user releases the recording hotkey.
+type ReleaseTailCaptureEngine interface {
+ ReleaseTailCaptureDelay() time.Duration
+}
+
// StreamingSession receives incremental 16kHz mono PCM and returns the current
// best transcript for the active utterance.
type StreamingSession interface {
diff --git a/privatevoice.src/internal/engine/engine_darwin.go b/privatevoice.src/internal/engine/engine_darwin.go
index 3fc59e1..d530c83 100755
--- a/privatevoice.src/internal/engine/engine_darwin.go
+++ b/privatevoice.src/internal/engine/engine_darwin.go
@@ -6,6 +6,7 @@
"fmt"
"strings"
"sync"
+ "time"
"voicesnap/internal/logger"
"voicesnap/internal/model"
@@ -16,6 +17,7 @@
const (
asrSampleRate = 16000
xasrTailPaddingSamples = asrSampleRate + asrSampleRate/2
+ xasrReleaseTailDelay = 300 * time.Millisecond
)
var xasrTailPadding = make([]float32, xasrTailPaddingSamples)
@@ -301,6 +303,10 @@
return e.hwInfo
}
+func (e *xasrStreamingEngine) ReleaseTailCaptureDelay() time.Duration {
+ return xasrReleaseTailDelay
+}
+
func (e *sherpaEngine) Close() {
if e.recognizer != nil {
sherpa.DeleteOfflineRecognizer(e.recognizer)
diff --git a/privatevoice.src/internal/engine/engine_darwin_test.go b/privatevoice.src/internal/engine/engine_darwin_test.go
index fdfbf8e..a46bcea 100644
--- a/privatevoice.src/internal/engine/engine_darwin_test.go
+++ b/privatevoice.src/internal/engine/engine_darwin_test.go
@@ -6,6 +6,7 @@
"os"
"path/filepath"
"testing"
+ "time"
"voicesnap/internal/model"
)
@@ -163,6 +164,19 @@
}
}
+func TestXASRRequestsReleaseTailCaptureDelay(t *testing.T) {
+ got := (&xasrStreamingEngine{}).ReleaseTailCaptureDelay()
+ if got != 300*time.Millisecond {
+ t.Fatalf("ReleaseTailCaptureDelay() = %v, want 300ms", got)
+ }
+}
+
+func TestOfflineSherpaDoesNotRequestReleaseTailCapture(t *testing.T) {
+ if _, ok := any(&sherpaEngine{}).(ReleaseTailCaptureEngine); ok {
+ t.Fatal("offline sherpa engine should not request release tail capture")
+ }
+}
+
func TestXASRRealModelSmoke(t *testing.T) {
if os.Getenv("PRIVATEVOICE_XASR_SMOKE") != "1" {
t.Skip("set PRIVATEVOICE_XASR_SMOKE=1 to run the real X-ASR model smoke test")
--
Gitblit v1.9.3