From 2ba1056c56435f8d98110eca18dba90dc344de66 Mon Sep 17 00:00:00 2001
From: Ariver <shanghai3168@gmail.com>
Date: Wed, 01 Jul 2026 23:57:44 +0800
Subject: [PATCH] Polish Windows overlay preview
---
privatevoice.src/internal/engine/engine_darwin_test.go | 103 +++++++++++++++++++++++++++++++++++++--------------
1 files changed, 74 insertions(+), 29 deletions(-)
diff --git a/privatevoice.src/internal/engine/engine_darwin_test.go b/privatevoice.src/internal/engine/engine_darwin_test.go
index a46bcea..245adc7 100644
--- a/privatevoice.src/internal/engine/engine_darwin_test.go
+++ b/privatevoice.src/internal/engine/engine_darwin_test.go
@@ -52,22 +52,22 @@
}
}
-func TestOfflineConfigForQwen3ASRUsesQwen3Config(t *testing.T) {
+func TestOfflineConfigForZipformerUsesPlainTransducer(t *testing.T) {
root := t.TempDir()
files := map[string]string{
- "conv_frontend": filepath.Join(root, "conv_frontend.onnx"),
- "encoder": filepath.Join(root, "encoder.int8.onnx"),
- "decoder": filepath.Join(root, "decoder.int8.onnx"),
- "tokenizer": filepath.Join(root, "tokenizer"),
+ "encoder": filepath.Join(root, "encoder-epoch-99-avg-1.int8.onnx"),
+ "decoder": filepath.Join(root, "decoder-epoch-99-avg-1.onnx"),
+ "joiner": filepath.Join(root, "joiner-epoch-99-avg-1.int8.onnx"),
+ "tokens": filepath.Join(root, "tokens.txt"),
}
resolved := model.ResolvedModel{
- ModelID: model.Qwen3ASRModelID,
- BackendKind: model.BackendQwen3ASR,
+ ModelID: model.JapaneseModelID,
+ BackendKind: model.BackendTransducer,
Profile: model.ModelProfile{
- DisplayName: "Qwen3-ASR",
- BackendKind: model.BackendQwen3ASR,
+ DisplayName: "Zipformer Japanese",
+ BackendKind: model.BackendTransducer,
ProviderOrder: []string{"cpu"},
- NumThreads: 4,
+ NumThreads: 1,
},
Files: files,
}
@@ -76,26 +76,20 @@
if err != nil {
t.Fatal(err)
}
- if config.ModelConfig.Qwen3ASR.ConvFrontend != files["conv_frontend"] {
- t.Fatalf("conv frontend = %q, want %q", config.ModelConfig.Qwen3ASR.ConvFrontend, files["conv_frontend"])
+ if config.ModelConfig.Transducer.Encoder != files["encoder"] {
+ t.Fatalf("encoder = %q, want %q", config.ModelConfig.Transducer.Encoder, files["encoder"])
}
- if config.ModelConfig.Qwen3ASR.Encoder != files["encoder"] {
- t.Fatalf("encoder = %q, want %q", config.ModelConfig.Qwen3ASR.Encoder, files["encoder"])
+ if config.ModelConfig.Transducer.Decoder != files["decoder"] {
+ t.Fatalf("decoder = %q, want %q", config.ModelConfig.Transducer.Decoder, files["decoder"])
}
- if config.ModelConfig.Qwen3ASR.Decoder != files["decoder"] {
- t.Fatalf("decoder = %q, want %q", config.ModelConfig.Qwen3ASR.Decoder, files["decoder"])
+ if config.ModelConfig.Transducer.Joiner != files["joiner"] {
+ t.Fatalf("joiner = %q, want %q", config.ModelConfig.Transducer.Joiner, files["joiner"])
}
- if config.ModelConfig.Qwen3ASR.Tokenizer != files["tokenizer"] {
- t.Fatalf("tokenizer = %q, want %q", config.ModelConfig.Qwen3ASR.Tokenizer, files["tokenizer"])
+ if config.ModelConfig.Tokens != files["tokens"] {
+ t.Fatalf("tokens = %q, want %q", config.ModelConfig.Tokens, files["tokens"])
}
- if config.ModelConfig.Qwen3ASR.MaxTotalLen != 1024 {
- t.Fatalf("max total len = %d, want 1024", config.ModelConfig.Qwen3ASR.MaxTotalLen)
- }
- if config.ModelConfig.Qwen3ASR.MaxNewTokens != 256 {
- t.Fatalf("max new tokens = %d, want 256", config.ModelConfig.Qwen3ASR.MaxNewTokens)
- }
- if config.ModelConfig.Tokens != "" {
- t.Fatalf("tokens = %q, want empty for qwen3-asr", config.ModelConfig.Tokens)
+ if config.ModelConfig.ModelType != "" {
+ t.Fatalf("model type = %q, want empty for plain transducer", config.ModelConfig.ModelType)
}
}
@@ -164,6 +158,28 @@
}
}
+func TestXASRHeadPaddingPreparesInitialContext(t *testing.T) {
+ if xasrHeadPaddingSamples != 4000 {
+ t.Fatalf("head padding samples = %d, want 4000", xasrHeadPaddingSamples)
+ }
+ input := []float32{0.1, -0.2, 0.3}
+ got := xasrSamplesWithHeadPadding(input)
+
+ if len(got) != xasrHeadPaddingSamples+len(input) {
+ t.Fatalf("padded len = %d, want %d", len(got), xasrHeadPaddingSamples+len(input))
+ }
+ for i := 0; i < xasrHeadPaddingSamples; i++ {
+ if got[i] != 0 {
+ t.Fatalf("head padding sample %d = %v, want 0", i, got[i])
+ }
+ }
+ for i, want := range input {
+ if got[xasrHeadPaddingSamples+i] != want {
+ t.Fatalf("payload sample %d = %v, want %v", i, got[xasrHeadPaddingSamples+i], want)
+ }
+ }
+}
+
func TestXASRRequestsReleaseTailCaptureDelay(t *testing.T) {
got := (&xasrStreamingEngine{}).ReleaseTailCaptureDelay()
if got != 300*time.Millisecond {
@@ -171,9 +187,38 @@
}
}
-func TestOfflineSherpaDoesNotRequestReleaseTailCapture(t *testing.T) {
- if _, ok := any(&sherpaEngine{}).(ReleaseTailCaptureEngine); ok {
- t.Fatal("offline sherpa engine should not request release tail capture")
+func TestSenseVoiceRequestsReleaseTailCaptureDelay(t *testing.T) {
+ got := (&sherpaEngine{backendKind: model.BackendSenseVoice}).ReleaseTailCaptureDelay()
+ if got != 100*time.Millisecond {
+ t.Fatalf("ReleaseTailCaptureDelay() = %v, want 100ms", got)
+ }
+}
+
+func TestOtherSherpaModelsDoNotRequestReleaseTailCaptureDelay(t *testing.T) {
+ for _, backendKind := range []string{
+ model.BackendMoonshine,
+ model.BackendTransducer,
+ model.BackendNemoTransducer,
+ model.BackendQwen3ASR,
+ } {
+ t.Run(backendKind, func(t *testing.T) {
+ got := (&sherpaEngine{backendKind: backendKind}).ReleaseTailCaptureDelay()
+ if got != 0 {
+ t.Fatalf("ReleaseTailCaptureDelay() = %v, want 0", got)
+ }
+ })
+ }
+}
+
+func TestXASREnablesHoldPreCapture(t *testing.T) {
+ if !(&xasrStreamingEngine{}).HoldPreCaptureEnabled() {
+ t.Fatal("X-ASR should enable hold pre-capture")
+ }
+}
+
+func TestOfflineSherpaDoesNotEnableHoldPreCapture(t *testing.T) {
+ if _, ok := any(&sherpaEngine{}).(HoldPreCaptureEngine); ok {
+ t.Fatal("offline sherpa engine should not enable hold pre-capture")
}
}
--
Gitblit v1.9.3