From 50bec4fdb62e4848fbac89d6720828730ffd0b42 Mon Sep 17 00:00:00 2001 From: Rogee Date: Thu, 27 Aug 2026 22:05:28 +0800 Subject: [PATCH] feat: improve voice controls and interruption --- README.md | 14 ++-- asr_bailian.go | 40 +++++++--- asr_volc.go | 52 +++++++++---- config.go | 4 + llm.go | 2 + main.go | 136 +++++++++++++++++++++++++-------- tts.go | 67 ++++++++++++++--- tts_test.go | 80 ++++++++++++++++++++ voices.go | 76 +++++++++++++++++++ web/index.html | 201 ++++++++++++++++++++++++++++++++++++++++++++++--- 10 files changed, 589 insertions(+), 83 deletions(-) create mode 100644 tts_test.go create mode 100644 voices.go diff --git a/README.md b/README.md index ff805b5..e8c8086 100644 --- a/README.md +++ b/README.md @@ -1,6 +1,6 @@ # voice_test — 实时 ASR 客服机器人测试台 -Chrome 麦克风 → 实时 ASR（可选多模型）→ LLM 客服人设回答 → TTS 播报，全链路验证。 +Chrome 麦克风 → 实时 ASR（可选多模型）→ LLM 客服人设回答 → 下拉选择 TTS 模型、预置/复刻音色并配置语速、语调、音量和表达指令；页面还可录制并下载复刻用 WAV 样本。 ## 运行 @@ -39,7 +39,9 @@ export VOLC_ASR_RESOURCE_ID=volc.bigasr.sauc.duration # 2.0 用 volc.seedasr.sa | `BAILIAN_BASE_URL` / `BAILIAN_API_KEY` / `BAILIAN_WSS_BASE_URL` | - | 阿里百炼（zshenv 已有） | | `BAILIAN_LLM_MODEL` | `qwen3.8-flash` | 客服回答模型（OpenAI 兼容流式） | | `BAILIAN_TTS_MODEL` | `cosyvoice-v3.5-plus` | 语音合成模型 | -| `BAILIAN_TTS_VOICE` | zshenv 音色 | cosyvoice-v3.5-plus + 复刻音色已验证可用 | +| `BAILIAN_TTS_VOICE` | zshenv 音色 | 默认音色或百炼复刻/设计生成的 `voice_id` | +| `BAILIAN_TTS_INSTRUCTION` | 空 | 默认声音表达指令，网页可覆盖 | +| `BAILIAN_VOICE_API_URL` | DashScope 声音管理接口 | 查询当前账号的复刻/设计音色下拉列表 | | `PORT` | `:8090` | 监听地址 | ## 已验证的端到端指标（同一句“查询订单发货”，50ms/块推流，句级流式 TTS + 无 thinking） @@ -69,7 +71,8 @@ ASR 尾静音实测：语音结束→final 判定 ≈ 0.5s（qwen 0.6s），`max - `asr_volc.go` — 火山 sauc bigmodel（二进制帧协议） - `llm.go` — OpenAI 兼容流式对话 + 客服 system prompt - `tts.go` — CosyVoice 流式合成（PCM16/16k） -- `web/index.html` — 测试页（AudioWorklet 采集 PCM16/16k、音量条、双栏转录） +- `voices.go` — TTS 模型、预置音色与百炼复刻音色列表 +- `web/index.html` — 测试页（AudioWorklet 采集 PCM16/16k、WAV 样本录制、音量条、双栏转录） ## 已知简化 @@ -95,7 +98,7 @@ source ~/.zshenv cd ~/Workspace/voice_test && go build -o voicetest . && ./voicetest ``` -1. [ ] Chrome 打开 `http://localhost:8090`，状态栏显示“已连接”，下拉框列出 4 个模型 +1. [ ] Chrome 打开 `http://localhost:8090`，ASR/TTS 模型及预置/复刻音色均显示为下拉框 2. [ ] 选 `qwen-audio-3.0-asr-flash-streaming` → 🎤 开始 → 授权麦克风 3. [ ] 说“你好，我想查一下我的订单什么时候发货”： - [ ] 左栏灰色 partial 实时上屏，停顿后固化为 final @@ -105,7 +108,8 @@ cd ~/Workspace/voice_test && go build -o voicetest . && ./voicetest 4. [ ] 追问“订单号是 12345”：确认多轮上下文生效（回答应引用订单号） 5. [ ] 停止 → 重新开始，切换 `fun-asr-realtime` / `fun-asr-flash-2026-06-15` 重复 3 6. [ ] 点停止后 asr badge 回 idle；服务端日志出现 `ASR started` 而非报错 -7. [ ] 火山：选 `volc-bigmodel` 重复 3、4（已自动化验证通过，人工复测语音即可） +7. [ ] 点击“录制复刻样本”，录制后可试听并下载单声道 PCM WAV +8. [ ] 火山：选 `volc-bigmodel` 重复 3、4（已自动化验证通过，人工复测语音即可） ### 🧹 清理 diff --git a/asr_bailian.go b/asr_bailian.go index dd6e74b..f2daa20 100644 --- a/asr_bailian.go +++ b/asr_bailian.go @@ -5,6 +5,7 @@ import ( "fmt" "strings" "sync" + "sync/atomic" "github.com/gorilla/websocket" ) @@ -15,11 +16,13 @@ type bailianASR struct { model string api agentCfg - conn *websocket.Conn - taskID string - out chan asrEvent - wmu sync.Mutex // 上游连接写锁（sendAudio 与 updateContext 并发） - once sync.Once + conn *websocket.Conn + taskID string + out chan asrEvent + wmu sync.Mutex // 上游连接写锁（sendAudio 与 updateContext 并发） + connOnce sync.Once + outOnce sync.Once + stopping atomic.Bool } func newBailianASR(model string, api agentCfg) *bailianASR { @@ -33,6 +36,12 @@ func (a *bailianASR) start(ctx context.Context) error { return fmt.Errorf("连接百炼 ASR: %w", err) } a.conn = conn + started := false + defer func() { + if !started { + conn.Close() + } + }() a.taskID = uuid() err = a.writeJSON(map[string]any{ "header": map[string]string{"action": "run-task", "task_id": a.taskID, "streaming": "duplex"}, @@ -59,6 +68,7 @@ func (a *bailianASR) start(ctx context.Context) error { if ev.Header.Event != "task-started" { return fmt.Errorf("启动任务失败: %s %s", ev.Header.ErrorCode, ev.Header.ErrorMessage) } + started = true go a.readLoop() return nil } @@ -82,7 +92,9 @@ func (a *bailianASR) readLoop() { } `json:"payload"` } if err := a.conn.ReadJSON(&ev); err != nil { - a.emit(asrEvent{Typ: "error", Error: fmt.Sprintf("连接关闭: %v", err)}) + if !a.stopping.Load() { + a.emit(asrEvent{Typ: "error", Error: fmt.Sprintf("连接关闭: %v", err)}) + } return } switch ev.Header.Event { @@ -146,13 +158,21 @@ func (a *bailianASR) finish() error { } func (a *bailianASR) events() <-chan asrEvent { return a.out } -func (a *bailianASR) close() { a.closeUpstream() } +func (a *bailianASR) close() { + a.stopping.Store(true) + a.closeConn() +} -func (a *bailianASR) closeUpstream() { - a.once.Do(func() { +func (a *bailianASR) closeConn() { + a.connOnce.Do(func() { if a.conn != nil { a.conn.Close() } - close(a.out) }) } + +// 只有 readLoop 退出后才能关闭事件通道，避免 close() 与 emit() 并发导致 panic。 +func (a *bailianASR) closeUpstream() { + a.closeConn() + a.outOnce.Do(func() { close(a.out) }) +} diff --git a/asr_volc.go b/asr_volc.go index bd761f4..f2f9163 100644 --- a/asr_volc.go +++ b/asr_volc.go @@ -10,6 +10,7 @@ import ( "io" "net/http" "sync" + "sync/atomic" "github.com/gorilla/websocket" ) @@ -21,9 +22,9 @@ const ( volcURL = "wss://openspeech.bytedance.com/api/v3/sauc/bigmodel" msgFullClient = 0x1 - msgAudioOnly = 0x2 - msgFullServer = 0x9 - msgError = 0xF + msgAudioOnly = 0x2 + msgFullServer = 0x9 + msgError = 0xF flagNoSeq = 0x0 // full client request 用 flagLast = 0x2 // 客户端最后一包音频（负包，无序号） @@ -34,12 +35,14 @@ const ( ) type volcASR struct { - cfg agentCfg - conn *websocket.Conn - out chan asrEvent - wmu sync.Mutex - once sync.Once - emitted map[int]bool // 已作为 final 发出的 utterance 下标 + cfg agentCfg + conn *websocket.Conn + out chan asrEvent + wmu sync.Mutex + connOnce sync.Once + outOnce sync.Once + stopping atomic.Bool + emitted map[int]bool // 已作为 final 发出的 utterance 下标 } func newVolcASR(cfg agentCfg) *volcASR { @@ -70,10 +73,16 @@ func (v *volcASR) start(ctx context.Context) error { return fmt.Errorf("连接火山 ASR (401/403 多为凭证或资源未开通, logid=%s): %w", logid, err) } v.conn = conn + started := false + defer func() { + if !started { + conn.Close() + } + }() req := map[string]any{ - "user": map[string]any{"uid": "voice-test-agent"}, - "audio": map[string]any{"format": "pcm", "rate": 16000, "bits": 16, "channel": 1}, + "user": map[string]any{"uid": "voice-test-agent"}, + "audio": map[string]any{"format": "pcm", "rate": 16000, "bits": 16, "channel": 1}, "request": map[string]any{ "model_name": "bigmodel", "enable_itn": true, @@ -86,6 +95,7 @@ func (v *volcASR) start(ctx context.Context) error { if err := v.conn.WriteMessage(websocket.BinaryMessage, volcFrame(msgFullClient, flagNoSeq, serJSON, compNone, body)); err != nil { return fmt.Errorf("发送 full client request: %w", err) } + started = true go v.readLoop() return nil } @@ -107,7 +117,9 @@ func (v *volcASR) readLoop() { for { mt, data, err := v.conn.ReadMessage() if err != nil { - v.emit(asrEvent{Typ: "error", Error: fmt.Sprintf("连接关闭: %v", err)}) + if !v.stopping.Load() { + v.emit(asrEvent{Typ: "error", Error: fmt.Sprintf("连接关闭: %v", err)}) + } return } if mt != websocket.BinaryMessage || len(data) < 8 { @@ -205,13 +217,21 @@ func (v *volcASR) finish() error { } func (v *volcASR) events() <-chan asrEvent { return v.out } -func (v *volcASR) close() { v.closeUpstream() } +func (v *volcASR) close() { + v.stopping.Store(true) + v.closeConn() +} -func (v *volcASR) closeUpstream() { - v.once.Do(func() { +func (v *volcASR) closeConn() { + v.connOnce.Do(func() { if v.conn != nil { v.conn.Close() } - close(v.out) }) } + +// 只有 readLoop 退出后才能关闭事件通道，避免 close() 与 emit() 并发导致 panic。 +func (v *volcASR) closeUpstream() { + v.closeConn() + v.outOnce.Do(func() { close(v.out) }) +} diff --git a/config.go b/config.go index 6f9fafe..486e9da 100644 --- a/config.go +++ b/config.go @@ -12,6 +12,8 @@ type agentCfg struct { BailianKey string TTSModel string TTSVoice string + TTSInstruction string + VoiceAPIURL string LLMModel string VolcAppKey string // 新版控制台：单凭证 VolcAppID string // 旧版控制台：APP ID + Access Token 两件套 @@ -26,6 +28,8 @@ func loadCfg() agentCfg { BailianKey: getenv("BAILIAN_API_KEY", ""), TTSModel: getenv("BAILIAN_TTS_MODEL", "cosyvoice-v3.5-plus"), TTSVoice: getenv("BAILIAN_TTS_VOICE", ""), + TTSInstruction: getenv("BAILIAN_TTS_INSTRUCTION", ""), + VoiceAPIURL: getenv("BAILIAN_VOICE_API_URL", "https://dashscope.aliyuncs.com/api/v1/services/audio/tts/customization"), LLMModel: getenv("BAILIAN_LLM_MODEL", "qwen3.8-flash"), VolcAppKey: getenv("VOLC_ASR_APP_KEY", ""), VolcAppID: getenv("VOLC_ASR_APP_ID", ""), diff --git a/llm.go b/llm.go index 801911d..33914b7 100644 --- a/llm.go +++ b/llm.go @@ -29,6 +29,7 @@ func llmChat(ctx context.Context, cfg agentCfg, history []chatMsg) (<-chan strin req, err := http.NewRequestWithContext(ctx, "POST", cfg.BailianBaseURL+"/chat/completions", bytes.NewReader(body)) if err != nil { + close(textCh) errCh <- err return textCh, errCh } @@ -37,6 +38,7 @@ func llmChat(ctx context.Context, cfg agentCfg, history []chatMsg) (<-chan strin resp, err := http.DefaultClient.Do(req) if err != nil { + close(textCh) errCh <- err return textCh, errCh } diff --git a/main.go b/main.go index d1a4218..80ad87e 100644 --- a/main.go +++ b/main.go @@ -13,7 +13,7 @@ import ( ) var upgrader = websocket.Upgrader{ - CheckOrigin: func(*http.Request) bool { return true }, // 本地测试页，放开来源 + CheckOrigin: func(*http.Request) bool { return true }, // 本地测试页，放开来源 ReadBufferSize: 16384, WriteBufferSize: 16384, } @@ -30,9 +30,10 @@ func asrModels(cfg agentCfg) []map[string]any { } type clientMsg struct { - Type string `json:"type"` - ASR string `json:"asr"` - Mode string `json:"mode"` + Type string `json:"type"` + ASR string `json:"asr"` + Mode string `json:"mode"` + TTS *ttsCfg `json:"tts"` } type chatMsg struct { @@ -41,13 +42,16 @@ type chatMsg struct { } type session struct { - cfg agentCfg - conn *websocket.Conn - wmu sync.Mutex // 客户端连接写锁 - mu sync.Mutex // provider / history 状态锁 - prov asrProvider - hist []chatMsg - cancel context.CancelFunc + cfg agentCfg + conn *websocket.Conn + wmu sync.Mutex // 客户端连接写锁 + mu sync.Mutex // provider / history 状态锁 + prov asrProvider + tts ttsCfg + hist []chatMsg + cancel context.CancelFunc + turnCancel context.CancelFunc + turnID uint64 } func (s *session) sendJSON(v any) { @@ -73,12 +77,29 @@ func main() { http.HandleFunc("/api/models", func(w http.ResponseWriter, r *http.Request) { json.NewEncoder(w).Encode(asrModels(cfg)) }) + http.HandleFunc("/api/config", func(w http.ResponseWriter, r *http.Request) { + ctx, cancel := context.WithTimeout(r.Context(), 3*time.Second) + defer cancel() + cloned, voiceErr := listClonedVoices(ctx, cfg) + found := false + for _, v := range cloned { + found = found || v.ID == cfg.TTSVoice + } + if cfg.TTSVoice != "" && !found { + cloned = append([]voiceOption{{ID: cfg.TTSVoice, Label: cfg.TTSVoice + "（当前配置）"}}, cloned...) + } + payload := map[string]any{"tts": defaultTTS(cfg), "ttsModels": ttsModels(), "systemVoices": systemVoices(), "clonedVoices": cloned} + if voiceErr != nil { + payload["voiceError"] = voiceErr.Error() + } + json.NewEncoder(w).Encode(payload) + }) http.HandleFunc("/agent", func(w http.ResponseWriter, r *http.Request) { conn, err := upgrader.Upgrade(w, r, nil) if err != nil { return } - s := &session{cfg: cfg, conn: conn} + s := &session{cfg: cfg, conn: conn, tts: defaultTTS(cfg)} defer conn.Close() s.readLoop() }) @@ -102,14 +123,9 @@ func (s *session) readLoop() { } switch m.Type { case "start": - s.startASR(m.ASR) + s.startASR(m.ASR, m.TTS) case "stop": - s.mu.Lock() - p := s.prov - s.mu.Unlock() - if p != nil { - p.finish() - } + s.stopASR() } case websocket.BinaryMessage: s.mu.Lock() @@ -122,7 +138,12 @@ func (s *session) readLoop() { } } -func (s *session) startASR(id string) { +func (s *session) startASR(id string, requestedTTS *ttsCfg) { + tts, err := normalizeTTS(requestedTTS, s.cfg) + if err != nil { + s.sendJSON(map[string]any{"type": "error", "error": err.Error()}) + return + } s.stopASR() ctx, cancel := context.WithCancel(context.Background()) var p asrProvider @@ -138,13 +159,14 @@ func (s *session) startASR(id string) { return } s.mu.Lock() - s.prov, s.cancel = p, cancel + s.prov, s.cancel, s.tts = p, cancel, tts s.mu.Unlock() go s.eventPump(p) log.Printf("ASR started: %s", id) } func (s *session) stopASR() { + s.interruptTurn() s.mu.Lock() p, cancel := s.prov, s.cancel s.prov, s.cancel = nil, nil @@ -158,7 +180,38 @@ func (s *session) stopASR() { } } -// eventPump: ASR 事件 → 转发给浏览器；final 句触发客服回答（LLM 流式 → TTS 流式）。 +func (s *session) interruptTurn() { + s.mu.Lock() + cancel := s.turnCancel + s.turnCancel = nil + s.mu.Unlock() + if cancel != nil { + cancel() + if s.conn != nil { + s.sendJSON(map[string]string{"type": "interrupt"}) + } + } +} + +func (s *session) startTurn() (context.Context, context.CancelFunc, uint64) { + ctx, cancel := context.WithTimeout(context.Background(), 120*time.Second) + s.mu.Lock() + s.turnID++ + id := s.turnID + s.turnCancel = cancel + s.mu.Unlock() + return ctx, cancel, id +} + +func (s *session) finishTurn(id uint64) { + s.mu.Lock() + if s.turnID == id { + s.turnCancel = nil + } + s.mu.Unlock() +} + +// eventPump: 新语音一出现即打断当前回答；final 句异步触发下一轮，保持 ASR 事件可继续消费。 func (s *session) eventPump(p asrProvider) { defer s.sendJSON(map[string]string{"type": "asr-stopped"}) for ev := range p.events() { @@ -166,21 +219,24 @@ func (s *session) eventPump(p asrProvider) { case "error": s.sendJSON(map[string]any{"type": "error", "error": ev.Error, "code": ev.Code}) case "partial", "final": + s.interruptTurn() s.sendJSON(map[string]any{"type": ev.Typ, "text": ev.Text}) if ev.Typ == "final" { - s.agentTurn(p, ev.Text) + ctx, cancel, id := s.startTurn() + go s.agentTurn(ctx, cancel, id, p, ev.Text) } } } } // agentTurn: final 句 → LLM 流式出字（逐段回传）→ 边生成边按句喂 TTS → 音频流回传。 -func (s *session) agentTurn(p asrProvider, userText string) { - ctx, cancel := context.WithTimeout(context.Background(), 120*time.Second) +func (s *session) agentTurn(ctx context.Context, cancel context.CancelFunc, id uint64, p asrProvider, userText string) { defer cancel() + defer s.finishTurn(id) s.mu.Lock() hist := append([]chatMsg(nil), s.hist...) + tts := s.tts s.mu.Unlock() hist = append(hist, chatMsg{Role: "user", Content: userText}) @@ -190,7 +246,7 @@ func (s *session) agentTurn(p asrProvider, userText string) { go func() { defer close(ttsDone) for sen := range sentences { - if err := ttsSpeak(ctx, s.cfg, sen, s.sendAudio); err != nil { + if err := ttsSpeak(ctx, s.cfg, tts, sen, s.sendAudio); err != nil && ctx.Err() == nil { s.sendJSON(map[string]any{"type": "error", "error": err.Error()}) } } @@ -199,27 +255,45 @@ func (s *session) agentTurn(p asrProvider, userText string) { textCh, errCh := llmChat(ctx, s.cfg, hist) var full []byte var pending []rune - flush := func() { + flush := func() bool { if sen := strings.TrimSpace(string(pending)); sen != "" { - sentences <- sen + select { + case sentences <- sen: + case <-ctx.Done(): + return false + } } pending = pending[:0] + return true } for seg := range textCh { + if ctx.Err() != nil { + continue // drain producer so cancellation cannot leave it blocked on textCh + } full = append(full, seg...) s.sendJSON(map[string]any{"type": "reply-delta", "text": seg}) for _, r := range seg { pending = append(pending, r) // 句末标点即切句；超长无标点时退到逗号，避免一句卡住整段 if strings.ContainsRune("。！？!?；;\n", r) || (len(pending) >= 100 && r == '，') { - flush() + if !flush() { + break + } } } } - flush() + if ctx.Err() == nil { + flush() + } close(sentences) if err := <-errCh; err != nil { - s.sendJSON(map[string]any{"type": "error", "error": "LLM: " + err.Error()}) + if ctx.Err() == nil { + s.sendJSON(map[string]any{"type": "error", "error": "LLM: " + err.Error()}) + } + <-ttsDone + return + } + if ctx.Err() != nil { <-ttsDone return } diff --git a/tts.go b/tts.go index 32ffe9c..46ea416 100644 --- a/tts.go +++ b/tts.go @@ -4,14 +4,52 @@ import ( "context" "encoding/json" "fmt" + "strings" "sync" "time" "github.com/gorilla/websocket" ) +type ttsCfg struct { + Model string `json:"model"` + Voice string `json:"voice"` + Instruction string `json:"instruction"` + Rate float64 `json:"rate"` + Pitch float64 `json:"pitch"` + Volume int `json:"volume"` +} + +func defaultTTS(cfg agentCfg) ttsCfg { + return ttsCfg{Model: cfg.TTSModel, Voice: cfg.TTSVoice, Instruction: cfg.TTSInstruction, Rate: 1, Pitch: 1, Volume: 50} +} + +func normalizeTTS(in *ttsCfg, cfg agentCfg) (ttsCfg, error) { + if in == nil { + return defaultTTS(cfg), nil + } + t := *in + t.Model, t.Voice, t.Instruction = strings.TrimSpace(t.Model), strings.TrimSpace(t.Voice), strings.TrimSpace(t.Instruction) + if t.Model == "" { + t.Model = cfg.TTSModel + } + if t.Voice == "" { + t.Voice = cfg.TTSVoice + } + if t.Voice == "" { + return t, fmt.Errorf("请填写音色或复刻音色 ID") + } + if len(t.Model) > 100 || len(t.Voice) > 200 || len([]rune(t.Instruction)) > 100 { + return t, fmt.Errorf("TTS 配置文本过长") + } + if t.Rate < 0.5 || t.Rate > 2 || t.Pitch < 0.5 || t.Pitch > 2 || t.Volume < 0 || t.Volume > 100 { + return t, fmt.Errorf("TTS 参数超出范围：语速/语调 0.5~2.0，音量 0~100") + } + return t, nil +} + // ttsSpeak: 百炼 CosyVoice 流式合成，音频块实时回调 onAudio。 -func ttsSpeak(ctx context.Context, cfg agentCfg, text string, onAudio func([]byte)) error { +func ttsSpeak(ctx context.Context, cfg agentCfg, tts ttsCfg, text string, onAudio func([]byte)) error { cctx, cancel := context.WithTimeout(ctx, 60*time.Second) defer cancel() conn, _, err := websocket.DefaultDialer.DialContext(cctx, cfg.BailianWssBaseURL, @@ -22,16 +60,21 @@ func ttsSpeak(ctx context.Context, cfg agentCfg, text string, onAudio func([]byt defer conn.Close() taskID := uuid() + parameters := map[string]any{ + "text_type": "PlainText", "voice": tts.Voice, + "format": "pcm", "sample_rate": 16000, + "volume": tts.Volume, "rate": tts.Rate, "pitch": tts.Pitch, + } + if tts.Instruction != "" { + parameters["instruction"] = tts.Instruction + } err = conn.WriteJSON(map[string]any{ "header": map[string]string{"action": "run-task", "task_id": taskID, "streaming": "duplex"}, "payload": map[string]any{ "task_group": "audio", "task": "tts", "function": "SpeechSynthesizer", - "model": cfg.TTSModel, - "parameters": map[string]any{ - "text_type": "PlainText", "voice": cfg.TTSVoice, - "format": "pcm", "sample_rate": 16000, - }, - "input": map[string]any{}, + "model": tts.Model, + "parameters": parameters, + "input": map[string]any{}, }, }) if err != nil { @@ -40,7 +83,6 @@ func ttsSpeak(ctx context.Context, cfg agentCfg, text string, onAudio func([]byt var once sync.Once done := make(chan error, 1) - var buf []json.RawMessage // task-started / task-finished 事件 go func() { for { mt, data, err := conn.ReadMessage() @@ -66,7 +108,6 @@ func ttsSpeak(ctx context.Context, cfg agentCfg, text string, onAudio func([]byt } switch ev.Header.Event { case "task-started": - buf = append(buf, json.RawMessage("{}")) once.Do(func() { // task-started 到达后才允许发文本 if err := conn.WriteJSON(map[string]any{ @@ -98,5 +139,11 @@ func ttsSpeak(ctx context.Context, cfg agentCfg, text string, onAudio func([]byt } } }() - return <-done + select { + case err := <-done: + return err + case <-cctx.Done(): + conn.Close() + return cctx.Err() + } } diff --git a/tts_test.go b/tts_test.go new file mode 100644 index 0000000..ea6f95d --- /dev/null +++ b/tts_test.go @@ -0,0 +1,80 @@ +package main + +import ( + "context" + "net/http" + "net/http/httptest" + "testing" +) + +func TestNormalizeTTS(t *testing.T) { + cfg := agentCfg{TTSModel: "cosyvoice-v3.5-plus", TTSVoice: "clone-id"} + + got, err := normalizeTTS(nil, cfg) + if err != nil || got.Voice != "clone-id" || got.Rate != 1 || got.Pitch != 1 || got.Volume != 50 { + t.Fatalf("defaults: got=%+v err=%v", got, err) + } + + bad := ttsCfg{Model: "m", Voice: "v", Rate: 2.1, Pitch: 1, Volume: 50} + if _, err := normalizeTTS(&bad, cfg); err == nil { + t.Fatal("expected out-of-range rate to fail") + } +} + +func TestASRCloseLeavesEventsForReader(t *testing.T) { + t.Run("bailian", func(t *testing.T) { + a := newBailianASR("model", agentCfg{}) + a.close() + a.emit(asrEvent{Typ: "final", Text: "ok"}) + if ev := <-a.events(); ev.Text != "ok" { + t.Fatalf("event=%+v", ev) + } + a.closeUpstream() + }) + t.Run("volc", func(t *testing.T) { + v := newVolcASR(agentCfg{}) + v.close() + v.emit(asrEvent{Typ: "final", Text: "ok"}) + if ev := <-v.events(); ev.Text != "ok" { + t.Fatalf("event=%+v", ev) + } + v.closeUpstream() + }) +} + +func TestLLMRequestErrorClosesStream(t *testing.T) { + text, errs := llmChat(context.Background(), agentCfg{BailianBaseURL: "://"}, nil) + if _, open := <-text; open { + t.Fatal("text stream left open after request error") + } + if err := <-errs; err == nil { + t.Fatal("expected request error") + } +} + +func TestInterruptTurn(t *testing.T) { + s := &session{} + ctx, cancel, _ := s.startTurn() + defer cancel() + s.interruptTurn() + select { + case <-ctx.Done(): + default: + t.Fatal("active turn was not canceled") + } +} + +func TestListClonedVoices(t *testing.T) { + server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) { + if r.Header.Get("Authorization") != "Bearer key" { + t.Fatal("missing authorization") + } + w.Write([]byte(`{"output":{"voice_list":[{"voice_id":"clone-ok","status":"OK"},{"voice_id":"clone-building","status":"DEPLOYING"}]}}`)) + })) + defer server.Close() + + voices, err := listClonedVoices(context.Background(), agentCfg{BailianKey: "key", VoiceAPIURL: server.URL}) + if err != nil || len(voices) != 1 || voices[0].ID != "clone-ok" { + t.Fatalf("voices=%+v err=%v", voices, err) + } +} diff --git a/voices.go b/voices.go new file mode 100644 index 0000000..33be11f --- /dev/null +++ b/voices.go @@ -0,0 +1,76 @@ +package main + +import ( + "bytes" + "context" + "encoding/json" + "fmt" + "net/http" +) + +type voiceOption struct { + ID string `json:"id"` + Label string `json:"label"` +} + +func ttsModels() []voiceOption { + return []voiceOption{ + {ID: "cosyvoice-v3.5-plus", Label: "CosyVoice v3.5 Plus（复刻/设计音色）"}, + {ID: "cosyvoice-v3.5-flash", Label: "CosyVoice v3.5 Flash（低延迟）"}, + {ID: "cosyvoice-v3-plus", Label: "CosyVoice v3 Plus"}, + {ID: "cosyvoice-v3-flash", Label: "CosyVoice v3 Flash"}, + {ID: "cosyvoice-v2", Label: "CosyVoice v2"}, + } +} + +func systemVoices() map[string][]voiceOption { + v3 := []voiceOption{{ID: "longanhuan", Label: "龙安欢｜欢脱元气女"}, {ID: "longanyang", Label: "龙安洋｜阳光男声"}} + return map[string][]voiceOption{ + "cosyvoice-v3-plus": v3, + "cosyvoice-v3-flash": v3, + "cosyvoice-v2": { + {ID: "longxiaochun_v2", Label: "龙小淳｜温柔女声"}, + {ID: "longwan_v2", Label: "龙婉｜普通话女声"}, + {ID: "loongbella_v2", Label: "Bella 2.0｜新闻女声"}, + }, + } +} + +func listClonedVoices(ctx context.Context, cfg agentCfg) ([]voiceOption, error) { + body, _ := json.Marshal(map[string]any{ + "model": "voice-enrollment", + "input": map[string]any{"action": "list_voice", "page_index": 0, "page_size": 100}, + }) + req, err := http.NewRequestWithContext(ctx, http.MethodPost, cfg.VoiceAPIURL, bytes.NewReader(body)) + if err != nil { + return nil, err + } + req.Header.Set("Authorization", "Bearer "+cfg.BailianKey) + req.Header.Set("Content-Type", "application/json") + resp, err := http.DefaultClient.Do(req) + if err != nil { + return nil, err + } + defer resp.Body.Close() + if resp.StatusCode/100 != 2 { + return nil, fmt.Errorf("查询复刻音色失败: HTTP %d", resp.StatusCode) + } + var result struct { + Output struct { + Voices []struct { + ID string `json:"voice_id"` + Status string `json:"status"` + } `json:"voice_list"` + } `json:"output"` + } + if err := json.NewDecoder(resp.Body).Decode(&result); err != nil { + return nil, err + } + voices := make([]voiceOption, 0, len(result.Output.Voices)) + for _, v := range result.Output.Voices { + if v.ID != "" && (v.Status == "" || v.Status == "OK" || v.Status == "SUCCESS") { + voices = append(voices, voiceOption{ID: v.ID, Label: v.ID}) + } + } + return voices, nil +} diff --git a/web/index.html b/web/index.html index 02f3558..4f0518a 100644 --- a/web/index.html +++ b/web/index.html @@ -7,24 +7,40 @@