From 3803f0987d7fd38cbd754f3dc37caa29247c852c Mon Sep 17 00:00:00 2001
From: Ariver <shanghai3168@gmail.com>
Date: Wed, 01 Jul 2026 23:21:27 +0800
Subject: [PATCH] Merge Windows preview build

---
 privatevoice.src/internal/engine/engine_darwin.go |   69 +++++++++++++++++++++++++---------
 1 files changed, 51 insertions(+), 18 deletions(-)

diff --git a/privatevoice.src/internal/engine/engine_darwin.go b/privatevoice.src/internal/engine/engine_darwin.go
index 3fc59e1..ddf1ad8 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"
 
@@ -15,14 +16,20 @@
 
 const (
 	asrSampleRate          = 16000
+	xasrHeadPaddingSamples = asrSampleRate / 4
 	xasrTailPaddingSamples = asrSampleRate + asrSampleRate/2
+	xasrReleaseTailDelay   = 300 * time.Millisecond
 )
 
-var xasrTailPadding = make([]float32, xasrTailPaddingSamples)
+var (
+	xasrHeadPadding = make([]float32, xasrHeadPaddingSamples)
+	xasrTailPadding = make([]float32, xasrTailPaddingSamples)
+)
 
 type sherpaEngine struct {
-	recognizer *sherpa.OfflineRecognizer
-	hwInfo     string
+	recognizer  *sherpa.OfflineRecognizer
+	backendKind string
+	hwInfo      string
 }
 
 type xasrStreamingEngine struct {
@@ -32,10 +39,11 @@
 }
 
 type xasrStreamingSession struct {
-	engine   *xasrStreamingEngine
-	stream   *sherpa.OnlineStream
-	lastText string
-	finished bool
+	engine             *xasrStreamingEngine
+	stream             *sherpa.OnlineStream
+	lastText           string
+	hasAcceptedSamples bool
+	finished           bool
 }
 
 func newPlatformEngine(resolved model.ResolvedModel) (Engine, error) {
@@ -59,8 +67,9 @@
 			info := fmt.Sprintf("%s · %s", resolved.Profile.DisplayName, p.name)
 			logger.Info("Engine initialized: %s", info)
 			return &sherpaEngine{
-				recognizer: recognizer,
-				hwInfo:     info,
+				recognizer:  recognizer,
+				backendKind: resolved.BackendKind,
+				hwInfo:      info,
 			}, nil
 		}
 		logger.Info("Failed to init with %s, trying next provider", p.name)
@@ -111,21 +120,19 @@
 		config.ModelConfig.Moonshine.Encoder = resolved.Files["encoder"]
 		config.ModelConfig.Moonshine.UncachedDecoder = resolved.Files["uncached_decoder"]
 		config.ModelConfig.Moonshine.CachedDecoder = resolved.Files["cached_decoder"]
+	case model.BackendTransducer:
+		config.ModelConfig.Transducer.Encoder = resolved.Files["encoder"]
+		config.ModelConfig.Transducer.Decoder = resolved.Files["decoder"]
+		config.ModelConfig.Transducer.Joiner = resolved.Files["joiner"]
 	case model.BackendNemoTransducer:
 		config.ModelConfig.Transducer.Encoder = resolved.Files["encoder"]
 		config.ModelConfig.Transducer.Decoder = resolved.Files["decoder"]
 		config.ModelConfig.Transducer.Joiner = resolved.Files["joiner"]
 		config.ModelConfig.ModelType = model.BackendNemoTransducer
 	case model.BackendQwen3ASR:
-		config.ModelConfig.Qwen3ASR.ConvFrontend = resolved.Files["conv_frontend"]
-		config.ModelConfig.Qwen3ASR.Encoder = resolved.Files["encoder"]
-		config.ModelConfig.Qwen3ASR.Decoder = resolved.Files["decoder"]
-		config.ModelConfig.Qwen3ASR.Tokenizer = resolved.Files["tokenizer"]
-		config.ModelConfig.Qwen3ASR.MaxTotalLen = 1024
-		config.ModelConfig.Qwen3ASR.MaxNewTokens = 256
-		config.ModelConfig.Qwen3ASR.Temperature = 0.0
-		config.ModelConfig.Qwen3ASR.TopP = 0.9
-		config.ModelConfig.Qwen3ASR.Seed = 0
+		if err := applyQwen3ASRConfig(&config, resolved); err != nil {
+			return sherpa.OfflineRecognizerConfig{}, err
+		}
 	default:
 		return sherpa.OfflineRecognizerConfig{}, fmt.Errorf("unsupported backend kind: %s", resolved.BackendKind)
 	}
@@ -233,6 +240,10 @@
 	if s.stream == nil || s.finished {
 		return s.lastText, nil
 	}
+	if !s.hasAcceptedSamples {
+		samples = xasrSamplesWithHeadPadding(samples)
+		s.hasAcceptedSamples = true
+	}
 
 	s.engine.mu.Lock()
 	defer s.engine.mu.Unlock()
@@ -293,14 +304,36 @@
 	return nextText
 }
 
+func xasrSamplesWithHeadPadding(samples []float32) []float32 {
+	padded := make([]float32, 0, len(xasrHeadPadding)+len(samples))
+	padded = append(padded, xasrHeadPadding...)
+	padded = append(padded, samples...)
+	return padded
+}
+
 func (e *sherpaEngine) HardwareInfo() string {
 	return e.hwInfo
 }
 
+func (e *sherpaEngine) ReleaseTailCaptureDelay() time.Duration {
+	if e.backendKind == model.BackendSenseVoice {
+		return senseVoiceReleaseTailDelay
+	}
+	return 0
+}
+
 func (e *xasrStreamingEngine) HardwareInfo() string {
 	return e.hwInfo
 }
 
+func (e *xasrStreamingEngine) ReleaseTailCaptureDelay() time.Duration {
+	return xasrReleaseTailDelay
+}
+
+func (e *xasrStreamingEngine) HoldPreCaptureEnabled() bool {
+	return true
+}
+
 func (e *sherpaEngine) Close() {
 	if e.recognizer != nil {
 		sherpa.DeleteOfflineRecognizer(e.recognizer)

--
Gitblit v1.9.3