From 0ffbf1935c9d091cce22a5583275ac0902f7a693 Mon Sep 17 00:00:00 2001
From: Ariver <shanghai3168@gmail.com>
Date: Thu, 02 Jul 2026 03:04:51 +0800
Subject: [PATCH] Add Cantonese language model support
---
privatevoice.src/internal/model/registry.go | 366 ++++++++++++++++++++++++++++++++++++++++++++++++++-
1 files changed, 353 insertions(+), 13 deletions(-)
diff --git a/privatevoice.src/internal/model/registry.go b/privatevoice.src/internal/model/registry.go
index 3d305c3..641adc4 100644
--- a/privatevoice.src/internal/model/registry.go
+++ b/privatevoice.src/internal/model/registry.go
@@ -8,8 +8,24 @@
const (
BackendSenseVoice = "sensevoice"
BackendMoonshine = "moonshine"
+ BackendTransducer = "transducer"
BackendNemoTransducer = "nemo_transducer"
+ BackendQwen3ASR = "qwen3_asr"
+ BackendXASRStreaming = "x_asr_streaming"
)
+
+var languageProfileOrder = []string{
+ DefaultLanguageID,
+ CantoneseLanguageID,
+ EnglishLanguageID,
+ FrenchLanguageID,
+ GermanLanguageID,
+ SpanishLanguageID,
+ ItalianLanguageID,
+ PortugueseLanguageID,
+ JapaneseLanguageID,
+ KoreanLanguageID,
+}
var modelProfiles = map[string]ModelProfile{
DefaultModelID: {
@@ -130,16 +146,264 @@
ProviderOrder: []string{"cpu"},
NumThreads: 4,
},
+ ParakeetV3ModelID: {
+ ID: ParakeetV3ModelID,
+ DisplayName: "Parakeet TDT v3",
+ BackendKind: BackendNemoTransducer,
+ Tier: "default",
+ SupportedLanguageIDs: []string{
+ FrenchLanguageID,
+ GermanLanguageID,
+ SpanishLanguageID,
+ ItalianLanguageID,
+ PortugueseLanguageID,
+ },
+ RecommendedFor: []string{
+ FrenchLanguageID,
+ GermanLanguageID,
+ SpanishLanguageID,
+ ItalianLanguageID,
+ PortugueseLanguageID,
+ },
+ Description: "欧洲多语种离线模型，适合法语、德语、西班牙语、意大利语和葡萄牙语听写。",
+ ApproxSize: "约 640 MiB 解压后模型目录",
+ DownloadURLs: []string{
+ "https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-nemo-parakeet-tdt-0.6b-v3-int8.tar.bz2",
+ },
+ InstallDirName: "parakeet-tdt-0.6b-v3",
+ RequiredFiles: []RequiredFileRule{
+ {
+ Role: "encoder",
+ AllOf: []string{"encoder.int8.onnx"},
+ Required: true,
+ },
+ {
+ Role: "decoder",
+ AllOf: []string{"decoder.int8.onnx"},
+ Required: true,
+ },
+ {
+ Role: "joiner",
+ AllOf: []string{"joiner.int8.onnx"},
+ Required: true,
+ },
+ {
+ Role: "tokens",
+ AllOf: []string{"tokens.txt"},
+ Required: true,
+ },
+ },
+ ProviderOrder: []string{"cpu"},
+ NumThreads: 2,
+ },
+ JapaneseModelID: {
+ ID: JapaneseModelID,
+ DisplayName: "Zipformer Japanese",
+ BackendKind: BackendTransducer,
+ Tier: "default",
+ SupportedLanguageIDs: []string{JapaneseLanguageID},
+ RecommendedFor: []string{JapaneseLanguageID},
+ Description: "日语专项离线模型，基于 ReazonSpeech 日语语料。",
+ ApproxSize: "约 165 MiB int8 关键模型文件，解压目录包含 fp32 对照文件",
+ DownloadURLs: []string{
+ "https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-zipformer-ja-reazonspeech-2024-08-01.tar.bz2",
+ },
+ InstallDirName: "zipformer-ja-reazonspeech",
+ RequiredFiles: []RequiredFileRule{
+ {
+ Role: "encoder",
+ AllOf: []string{"encoder-epoch-99-avg-1.int8.onnx"},
+ Required: true,
+ },
+ {
+ Role: "decoder",
+ AllOf: []string{"decoder-epoch-99-avg-1.onnx"},
+ Required: true,
+ },
+ {
+ Role: "joiner",
+ AllOf: []string{"joiner-epoch-99-avg-1.int8.onnx"},
+ Required: true,
+ },
+ {
+ Role: "tokens",
+ AllOf: []string{"tokens.txt"},
+ Required: true,
+ },
+ },
+ ProviderOrder: []string{"cpu"},
+ NumThreads: 1,
+ },
+ KoreanModelID: {
+ ID: KoreanModelID,
+ DisplayName: "Zipformer Korean",
+ BackendKind: BackendTransducer,
+ Tier: "default",
+ SupportedLanguageIDs: []string{KoreanLanguageID},
+ RecommendedFor: []string{KoreanLanguageID},
+ Description: "韩语专项离线模型，适合韩语听写。",
+ ApproxSize: "约 73 MiB int8 关键模型文件，解压目录包含 fp32 对照文件",
+ DownloadURLs: []string{
+ "https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-zipformer-korean-2024-06-24.tar.bz2",
+ },
+ InstallDirName: "zipformer-ko",
+ RequiredFiles: []RequiredFileRule{
+ {
+ Role: "encoder",
+ AllOf: []string{"encoder-epoch-99-avg-1.int8.onnx"},
+ Required: true,
+ },
+ {
+ Role: "decoder",
+ AllOf: []string{"decoder-epoch-99-avg-1.onnx"},
+ Required: true,
+ },
+ {
+ Role: "joiner",
+ AllOf: []string{"joiner-epoch-99-avg-1.int8.onnx"},
+ Required: true,
+ },
+ {
+ Role: "tokens",
+ AllOf: []string{"tokens.txt"},
+ Required: true,
+ },
+ },
+ ProviderOrder: []string{"cpu"},
+ NumThreads: 2,
+ },
+ Qwen3ASRModelID: {
+ ID: Qwen3ASRModelID,
+ DisplayName: "Qwen3-ASR",
+ BackendKind: BackendQwen3ASR,
+ Tier: "advanced",
+ SupportedLanguageIDs: []string{DefaultLanguageID, "zh-Hans", "zh-Hant", "yue"},
+ RecommendedFor: []string{DefaultLanguageID},
+ Description: "高质量中文离线模型，适合中文长句、技术词和中英混输准确率需求。",
+ ApproxSize: "约 838 MiB 下载包，解压后约 1.9 GiB",
+ DownloadURLs: []string{
+ "https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-qwen3-asr-0.6B-int8-2026-03-25.tar.bz2",
+ },
+ InstallDirName: "qwen3-asr-0.6b",
+ RequiredFiles: []RequiredFileRule{
+ {
+ Role: "conv_frontend",
+ AllOf: []string{"conv_frontend.onnx"},
+ Required: true,
+ },
+ {
+ Role: "encoder",
+ AllOf: []string{"encoder.int8.onnx"},
+ Required: true,
+ },
+ {
+ Role: "decoder",
+ AllOf: []string{"decoder.int8.onnx"},
+ Required: true,
+ },
+ {
+ Role: "tokenizer",
+ AllOf: []string{"tokenizer"},
+ Required: true,
+ Directory: true,
+ },
+ {
+ Role: "tokenizer_merges",
+ AllOf: []string{"tokenizer/merges.txt"},
+ Required: true,
+ },
+ {
+ Role: "tokenizer_vocab",
+ AllOf: []string{"tokenizer/vocab.json"},
+ Required: true,
+ },
+ },
+ ProviderOrder: []string{"cpu"},
+ NumThreads: 4,
+ },
+ XASRZhEn960ModelID: {
+ ID: XASRZhEn960ModelID,
+ DisplayName: "X-ASR zh-en 960ms",
+ BackendKind: BackendXASRStreaming,
+ Tier: "experimental",
+ SupportedLanguageIDs: []string{DefaultLanguageID, "zh-Hans", "zh-Hant", "yue", EnglishLanguageID},
+ RecommendedFor: []string{DefaultLanguageID, EnglishLanguageID},
+ Description: "实验性中英流式模型，适合评估中英混输和未来边说边出字体验。",
+ ApproxSize: "单个 960 ms 模型目录约 586 MiB",
+ InstallDirName: "x-asr-zh-en-960ms",
+ RequiredFiles: []RequiredFileRule{
+ {
+ Role: "encoder",
+ AllOf: []string{"encoder-960ms.onnx"},
+ Required: true,
+ },
+ {
+ Role: "decoder",
+ AllOf: []string{"decoder-960ms.onnx"},
+ Required: true,
+ },
+ {
+ Role: "joiner",
+ AllOf: []string{"joiner-960ms.onnx"},
+ Required: true,
+ },
+ {
+ Role: "tokens",
+ AllOf: []string{"tokens.txt"},
+ Required: true,
+ },
+ },
+ ProviderOrder: []string{"cpu"},
+ NumThreads: 1,
+ },
+ SenseVoiceYueModelID: {
+ ID: SenseVoiceYueModelID,
+ DisplayName: "SenseVoice Yue",
+ BackendKind: BackendSenseVoice,
+ Tier: "default",
+ SupportedLanguageIDs: []string{CantoneseLanguageID},
+ RecommendedFor: []string{CantoneseLanguageID},
+ Description: "粤语专项离线模型，基于 SenseVoice 并使用粤语数据增强。",
+ ApproxSize: "约 226 MiB 模型文件；下载包大小以实际 HTTP HEAD 为准",
+ LanguageParam: "yue",
+ DownloadURLs: []string{
+ "https://github.com/k2-fsa/sherpa-onnx/releases/download/asr-models/sherpa-onnx-sense-voice-zh-en-ja-ko-yue-int8-2025-09-09.tar.bz2",
+ },
+ InstallDirName: "sensevoice-yue-2025-09-09",
+ RequiredFiles: []RequiredFileRule{
+ {
+ Role: "model",
+ AllOf: []string{"model.int8.onnx"},
+ Required: true,
+ },
+ {
+ Role: "tokens",
+ AllOf: []string{"tokens.txt"},
+ Required: true,
+ },
+ },
+ ProviderOrder: []string{"coreml", "cpu"},
+ NumThreads: 4,
+ },
}
var languageProfiles = map[string]LanguageProfile{
DefaultLanguageID: {
- ID: DefaultLanguageID,
- DisplayName: "Chinese",
- NativeName: "中文",
+ ID: DefaultLanguageID,
+ DisplayName: "Chinese",
+ NativeName: "中文",
+ UILocale: "zh",
+ SystemMatchers: []string{"zh", "zh-CN", "zh-Hans", "zh-Hant"},
+ DefaultModelID: DefaultModelID,
+ UpgradeModelIDs: []string{Qwen3ASRModelID, XASRZhEn960ModelID},
+ },
+ CantoneseLanguageID: {
+ ID: CantoneseLanguageID,
+ DisplayName: "Cantonese",
+ NativeName: "粤语",
UILocale: "zh",
- SystemMatchers: []string{"zh", "zh-CN", "zh-Hans", "zh-Hant"},
- DefaultModelID: DefaultModelID,
+ SystemMatchers: []string{"yue", "yue-HK"},
+ DefaultModelID: SenseVoiceYueModelID,
},
EnglishLanguageID: {
ID: EnglishLanguageID,
@@ -148,7 +412,63 @@
UILocale: "en",
SystemMatchers: []string{"en", "en-US", "en-GB", "en-AU", "en-CA"},
DefaultModelID: MoonshineModelID,
- UpgradeModelIDs: []string{ParakeetModelID},
+ UpgradeModelIDs: []string{ParakeetModelID, XASRZhEn960ModelID},
+ },
+ FrenchLanguageID: {
+ ID: FrenchLanguageID,
+ DisplayName: "French",
+ NativeName: "Français",
+ UILocale: "fr",
+ SystemMatchers: []string{"fr", "fr-FR", "fr-BE", "fr-CA", "fr-CH"},
+ DefaultModelID: ParakeetV3ModelID,
+ },
+ GermanLanguageID: {
+ ID: GermanLanguageID,
+ DisplayName: "German",
+ NativeName: "Deutsch",
+ UILocale: "de",
+ SystemMatchers: []string{"de", "de-DE", "de-AT", "de-CH"},
+ DefaultModelID: ParakeetV3ModelID,
+ },
+ SpanishLanguageID: {
+ ID: SpanishLanguageID,
+ DisplayName: "Spanish",
+ NativeName: "Español",
+ UILocale: "es",
+ SystemMatchers: []string{"es", "es-ES", "es-MX", "es-US", "es-AR"},
+ DefaultModelID: ParakeetV3ModelID,
+ },
+ ItalianLanguageID: {
+ ID: ItalianLanguageID,
+ DisplayName: "Italian",
+ NativeName: "Italiano",
+ UILocale: "it",
+ SystemMatchers: []string{"it", "it-IT", "it-CH"},
+ DefaultModelID: ParakeetV3ModelID,
+ },
+ PortugueseLanguageID: {
+ ID: PortugueseLanguageID,
+ DisplayName: "Portuguese",
+ NativeName: "Português",
+ UILocale: "pt",
+ SystemMatchers: []string{"pt", "pt-PT", "pt-BR"},
+ DefaultModelID: ParakeetV3ModelID,
+ },
+ JapaneseLanguageID: {
+ ID: JapaneseLanguageID,
+ DisplayName: "Japanese",
+ NativeName: "日本語",
+ UILocale: "ja",
+ SystemMatchers: []string{"ja", "ja-JP"},
+ DefaultModelID: JapaneseModelID,
+ },
+ KoreanLanguageID: {
+ ID: KoreanLanguageID,
+ DisplayName: "Korean",
+ NativeName: "한국어",
+ UILocale: "ko",
+ SystemMatchers: []string{"ko", "ko-KR"},
+ DefaultModelID: KoreanModelID,
},
}
@@ -192,22 +512,42 @@
func ListLanguageProfiles() []LanguageProfile {
profiles := make([]LanguageProfile, 0, len(languageProfiles))
- for _, profile := range languageProfiles {
- profiles = append(profiles, profile)
+ seen := make(map[string]bool, len(languageProfiles))
+ for _, id := range languageProfileOrder {
+ if profile, ok := languageProfiles[id]; ok {
+ profiles = append(profiles, profile)
+ seen[id] = true
+ }
+ }
+ for id, profile := range languageProfiles {
+ if !seen[id] {
+ profiles = append(profiles, profile)
+ }
}
sort.Slice(profiles, func(i, j int) bool {
- if profiles[i].ID == DefaultLanguageID {
- return true
- }
- if profiles[j].ID == DefaultLanguageID {
- return false
+ left := languageOrderIndex(profiles[i].ID)
+ right := languageOrderIndex(profiles[j].ID)
+ if left != right {
+ return left < right
}
return profiles[i].ID < profiles[j].ID
})
return profiles
}
+func languageOrderIndex(id string) int {
+ for i, orderedID := range languageProfileOrder {
+ if id == orderedID {
+ return i
+ }
+ }
+ return len(languageProfileOrder)
+}
+
func NormalizeLanguageID(id string) string {
+ if id == "yue" {
+ return CantoneseLanguageID
+ }
if _, err := GetLanguageProfile(id); err != nil {
return DefaultLanguageID
}
--
Gitblit v1.9.3