dsh-her 0.1.0 → 0.2.0

package/lib/voice/voice.js

@@ -5,20 +5,22 @@
  *   process, engine.ts). Its model is downloaded in the background on first
  *   use; until then she only reads what you type.
  * - Speaking: the service of the voice she uses (core/voices.ts): Microsoft
- *   Edge voices through the edge-tts tool (edge.ts, the default), or Alibaba
- *   Cloud Model Studio CosyVoice (cosyvoice.ts). When the service cannot
+ *   Edge voices through the edge-tts tool (edge.ts, the default), Alibaba
+ *   Cloud Model Studio CosyVoice (cosyvoice.ts), or Ming on this computer
+ *   (ming.ts; built-in voices or her own, with her moods). When the service cannot
  *   speak (tool not installed yet, no key, no quota, offline), a sentence goes
  *   out without audio and shows as text; a key added later is picked up
  *   without restarting.
  * - Cache: short sentences ("嗯。" "好呀。") are kept, so the most common
  *   answers start at once.
  */
-import { serviceOf, VOICE_MODEL } from "../core/index.js";
+import { mingSpeech, serviceOf, spokenNumbers, VOICE_MODEL } from "../core/index.js";
 import { LocalVoice } from "./engine.js";
 import { CloudAccountError, CosyVoice } from "./cosyvoice.js";
 import { EdgeVoice } from "./edge.js";
 import { readCloudKey } from "./cloud-key.js";
 import { fetchModel, missingModels, modelPath, MODELS } from "./models.js";
+import { MING_STEPS } from "./ming-voices.js";
 const CACHE_MAX_CHARS = 8;
 const CACHE_ENTRIES = 64;
 /** After "no key" or "no quota", try the cloud again at most this often (a key added later is picked up). */
@@ -31,18 +33,20 @@ export class Voice {
     keyCheckedAt = 0;
     cloud;
     edge;
+    ming;
     cache = new Map();
     modelsDir;
     config;
     logger;
-    /** `edgeDir`: where dsh-her installs its own edge-tts if none is found. */
-    constructor(modelsDir, edgeDir, config, logger) {
+    /** `edgeDir`: where dsh-her installs its own edge-tts if none is found. `ming`: the local Ming voice and her own voices. */
+    constructor(modelsDir, edgeDir, config, logger, ming) {
         this.modelsDir = modelsDir;
         this.config = config;
         this.logger = logger;
         this.speakingState = config.enabled ? 'checking' : 'off';
         this.cloud = new CosyVoice(VOICE_MODEL, readCloudKey);
         this.edge = new EdgeVoice(edgeDir);
+        this.ming = ming;
     }
     /** Listening. */
     get status() {
@@ -50,6 +54,18 @@ export class Voice {
     }
     get speaking() {
         const service = serviceOf(this.config.voice);
+        if (service === 'ming' && this.config.enabled && this.ming !== undefined) {
+            // Ming tells its own state: installing, loading, stopped to save memory …
+            const ming = this.ming.voice.status;
+            if (ming.state === 'installing')
+                return { status: 'installing', service };
+            if (ming.state === 'loading')
+                return { status: 'checking', service };
+            if (ming.state === 'absent' || ming.state === 'unsupported')
+                return { status: 'failed', service, error: ming.error ?? '本机 Ming 还没安装' };
+            if (ming.state === 'failed')
+                return { status: 'failed', service, error: ming.error ?? '本机 Ming 出错了' };
+        }
         return this.speakingError === undefined ? { status: this.speakingState, service } : { status: this.speakingState, service, error: this.speakingError };
     }
     /** Check the cloud voice and get recognition ready. Never throws; the statuses say how it went. */
@@ -79,6 +95,8 @@ export class Voice {
     warm() {
         if (serviceOf(this.config.voice) === 'cosyvoice' && (this.speakingState === 'ready' || this.speakingState === 'failed'))
             this.cloud.warm();
+        if (serviceOf(this.config.voice) === 'ming')
+            this.ming?.voice.warm();
     }
     /** Change her voice and speed (from the "她" page); takes effect with the next sentence. */
     setVoice(voice, speed) {
@@ -96,18 +114,19 @@ export class Voice {
             throw new Error('她的声音已关闭（插件设置里 voice.enabled 是 false）');
         return this.speak(text, voice, speed);
     }
-    /** Her sentence as audio, or undefined when she cannot speak right now (it shows as text). */
-    async synthesize(text) {
+    /** Her sentence as audio, or undefined when she cannot speak right now (it shows as text). Its emotion colours a Ming voice. */
+    async synthesize(text, cues = []) {
         if (!this.config.enabled || text === '')
             return undefined;
         if ((this.speakingState === 'no-key' || this.speakingState === 'no-quota') && Date.now() - this.keyCheckedAt < KEY_RECHECK_MS)
             return undefined;
-        const key = `${this.config.voice}|${this.config.speed}|${text}`;
+        const emotion = serviceOf(this.config.voice) === 'ming' ? cues.find(c => c.kind === 'emotion')?.name : undefined;
+        const key = `${this.config.voice}|${this.config.speed}|${emotion ?? ''}|${text}`;
         const cached = this.cache.get(key);
         if (cached !== undefined)
             return cached;
         try {
-            const result = await this.speak(text);
+            const result = await this.speak(text, this.config.voice, this.config.speed, emotion);
             if ([...text].length <= CACHE_MAX_CHARS) {
                 this.cache.set(key, result);
                 if (this.cache.size > CACHE_ENTRIES)
@@ -134,11 +153,28 @@ export class Voice {
         this.asr?.stop();
         this.asr = undefined;
         this.cloud.close();
+        this.ming?.voice.close();
         if (this.listening !== 'off')
             this.listening = 'failed';
     }
-    async speak(text, voice = this.config.voice, speed = this.config.speed) {
+    async speak(text, voice = this.config.voice, speed = this.config.speed, emotion) {
         try {
+            if (serviceOf(voice) === 'ming') {
+                if (this.ming === undefined)
+                    throw new Error('本机 Ming 不可用');
+                const settings = this.ming.own.settings();
+                const how = mingSpeech(voice, settings.moods ? emotion : undefined, speed);
+                const ref = how.own === undefined ? undefined : this.ming.own.ref(how.own);
+                if (how.own !== undefined && ref === undefined)
+                    throw new Error('这个声音已经删掉了，换一个吧');
+                const audio = await this.ming.voice.synthesize({
+                    text: spokenNumbers(text), steps: MING_STEPS[settings.quality],
+                    ...(how.voice === undefined ? {} : { voice: how.voice }), ...(ref === undefined ? {} : { ref }),
+                    ...(how.emotion === undefined ? {} : { emotion: how.emotion }), ...(how.speed === undefined ? {} : { speed: how.speed }),
+                });
+                this.mark('ready');
+                return { audio: audio.audio, durationMs: audio.durationMs, format: 'wav' };
+            }
             if (serviceOf(voice) === 'edge') {
                 if (this.edge.status.status !== 'ready')
                     this.mark('installing');

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "dsh-her",
-  "version": "0.1.0",
+  "version": "0.2.0",
   "description": "A desktop companion for DeepSeek Harness (dsh): she lives on your real desktop (on top of your windows or inside the wallpaper), talks with you by voice and remembers you. · DSH 桌面女友：住在你真正的桌面上，能语音聊天、记得你。",
   "type": "module",
   "main": "lib/index.js",
@@ -23,6 +23,7 @@
   "author": "AIcivilization",
   "files": [
     "lib",
+    "python",
     "cordis.patch.yml",
     "README.md",
     "examples"

package/python/ming-requirements.in

@@ -0,0 +1 @@
+mlx-audio==0.6.3