arcane-os 0.49.0 → 0.50.0

package/docs/reference/ai/browser-speech.md

@@ -1,7 +1,7 @@
 # Browser speech providers
 
 `arcane-os/ai/browser-speech` supplies caller-selected Whisper or native Web
-Speech recognition, Kokoro text-to-speech, and TWiN Cloud text-to-speech providers. Its local path
+Speech recognition and synthesis, Kokoro text-to-speech, and TWiN Cloud text-to-speech providers. Its local path
 provides artifact storage, live module routing, role Workers, provider/2
 adapters, bounded parallel TTS synthesis, audio normalization, cancellation,
 and cleanup.
@@ -82,6 +82,78 @@ honestly rather than recording again or silently switching providers. See the
 [Web Speech specification](https://webaudio.github.io/web-speech-api/#speechreco-methods)
 for native stop/abort behavior.
 
+## Native browser text-to-speech and voice previews
+
+`createBrowserSpeechSynthesisProvider({id,model:{id,name?,defaultVoice?},language?})`
+uses the host browser's `speechSynthesis` and `SpeechSynthesisUtterance`.
+The application selects this provider as its default when desired; constructing
+it does not change global preferences, activate Chat speech, or play audio.
+The provider declares `localOnly:false` because a browser voice may use a remote
+service. Individual catalog records expose the browser's `localService` value.
+
+```javascript
+import {createBrowserSpeechSynthesisProvider} from 'arcane-os/ai/browser-speech';
+
+const voicePreview = createBrowserSpeechSynthesisProvider({
+  id: 'browser-preview',
+  model: {id: 'browser-voices', name: 'Browser voices'}
+});
+const unsubscribe = voicePreview.subscribeCatalog(function showVoices(models) {
+  // Pass these real records to the application's language/voice selector.
+  console.log(models[0].voices);
+});
+
+// Call only from the application's explicit Play action.
+function playIntroduction(input, voice) {
+  return voicePreview.play({input, voice});
+}
+// The owning view unsubscribes and disposes voicePreview when it closes.
+```
+
+`catalog()` synchronously returns an array of model records. Each model has
+`speech:{playback:'native'}` and `voices:[{id,name,lang,default,localService}]`;
+`id` is the exact native `voiceURI`. `subscribeCatalog(callback)` immediately
+replays that same shape and updates it on `voiceschanged`, returning an
+unsubscribe function. An initially empty inventory is ordinary pending browser
+state. There is no polling, discovery utterance, model download, or page-load
+barrier. Language selectors derive their choices from this actual inventory.
+
+`prepare({input,voice?,language?,speed?},{signal?})` returns an inert
+`{kind:'native-speech',input,voice,language,speed,play(options)}` descriptor.
+`play(payload,{signal,onState})` is the explicit-preview convenience. Both
+preserve complete input. Omitted voice uses the configured `model.defaultVoice`
+or the actual browser default; a supplied voice is never replaced with a named
+SDK alias. Language comes from that voice unless explicitly provided. Neither
+API reads the conversation's language preference. A saved voice waits for
+`voiceschanged` if the inventory is empty; a nonempty inventory missing it
+reports `ARCANE_AI_SPEECH_VOICE_UNAVAILABLE`.
+
+Playback returns `{finished,released,state,error,pause(),resume(),stop()}`.
+`finished` resolves `true` on actual native end, `false` on stop/abort, and
+rejects genuine native failures. `released` resolves when native resource
+ownership actually ends: a failed pause/resume/cancel can report an error before
+the utterance terminates. `onState({state,error})` observes queued,
+waiting-for-voices, playing, paused, complete, stopped, or error. Keep complete
+errors in diagnostics and ordinary status in the interface.
+
+Preview does not depend on AI mute or model activation. Use a separate provider
+instance when the preview should have a lifetime independent of Chat's selected
+provider. Provider unload/dispose stop their own playback, invalidate prepared
+descriptors, and disposal removes catalog listeners. Web Speech pause/resume/cancel
+operate on the native synthesis object globally; the SDK serializes its own
+utterances and does not claim control isolation from unrelated direct native
+callers. See the [native synthesis contract](https://webaudio.github.io/web-speech-api/#tts-section).
+
+For Chat, configure this provider with `ai.configureSpeechProvider('tts',provider)`
+and retain the existing explicit mute/unmute lifecycle. `AI.streamTTS()` and
+`SpeechPlayback` use actual utterance completion and cancellation in original
+order. `AI.prepareTTSPlayback(payload,signal)` returns either real Blob audio or
+the native descriptor without speaking; its signal remains attached by default
+when the returned descriptor is played. `AI.fetchTTS()` and durable
+`AI.prepareTTS()` remain audio-file APIs and report
+`ARCANE_AI_TTS_AUDIO_EXPORT_UNAVAILABLE` for native synthesis, which exposes no
+downloadable audio file.
+
 <a id="digitalocean-fal-text-to-speech"></a>
 
 ## TWiN Cloud text-to-speech

package/docs/reference/runtime-modules.md

@@ -160,7 +160,7 @@ read-only `ttsSegmentation`, `configureTTSSegmentation()`,
 `streamTTS(text='',end=false,options={})`,
 `prepareTTS({parts,storage,identity,signal,onState})`,
 `playPreparedTTS(prepared,{signal,onState})`,
-`finishTTS()`, `fetchTTS()`, `fetchSTT()`, `stopAudio()`, `resumeAudio()`,
+`finishTTS()`, `prepareTTSPlayback()`, `fetchTTS()`, `fetchSTT()`, `stopAudio()`, `resumeAudio()`,
 `playAudio()`; consumes `user-entity-loaded` and `arcane-ollama-ready`,
 installs `window.ai`, and emits `ai-ready` and `ai-tts-failure`.
 
@@ -374,7 +374,7 @@ marks from a cloned outbound input before delegation; the caller's payload stays
 unchanged. The third `preparation` argument is reserved for SDK-internal
 delegation, where `{speechInputPrepared:true}` prevents a second cleanup pass;
 applications omit it. `streamTTS(text='',end=false,options={})` and
-`finishTTS()` use this same request boundary. The third-argument options below
+`finishTTS()` use the selected playback boundary. The third-argument options below
 are available in SDK `0.5.12`. The `textFormat` compatibility extra added in
 SDK `0.5.16` is ignored beginning in `0.5.17`:
 
@@ -386,7 +386,7 @@ SDK `0.5.16` is ignored beginning in `0.5.17`:
 | `waitForPlayback` | `false` | Omission retains the preparation promise. With `true`, the promise resolves after every extracted segment reaches a terminal playback state: `true` when all naturally end, or `false` after terminal cancellation or failure. |
 | `textFormat` | Ignored compatibility extra | Repeated same formatting marks are removed automatically from every TTS call. This value no longer selects or disables cleanup. |
 
-The voice and speed use the existing `fetchTTS()` validation and error path.
+Audio-file voice and speed use the existing `fetchTTS()` validation and error path.
 The automatic cleanup changes only the outbound speech-input copy; displayed,
 stored, model, and caller-owned content remains exact. It is a narrow
 formatting-mark filter, not a full Markdown parser: links, code contents, list
@@ -405,6 +405,24 @@ for a complete passage. A call extracting no segments resolves
 `true` without waiting for earlier jobs; `finishTTS()` remains a preparation
 flush, not a queue-wide playback barrier. A muted call resolves `false`.
 
+`prepareTTSPlayback(payload,signal,preparation={})` is the silent playback-oriented
+request. Audio-file providers retain the `fetchTTS()` Blob contract; a selected
+model with `speech.playback:'native'` returns an inert `native-speech` descriptor.
+Its `play({signal,onState})` returns native completion and lifecycle controls.
+The preparation signal remains its default playback signal. Native input stays
+complete; per-request voice/language/speed and the configured model default own
+selection, without reading global user voice or conversation-language fields.
+Native streaming jobs wait for actual end in order and allocate no AudioContext.
+When native speech follows recorded audio, the preceding audio clock still owns
+its requested inter-part pause. Native trailing silence delays the next part,
+not the completed playback promise; cancellation retires either pending gap.
+Failed native control remains owned through its `released` promise, so Stop can
+still reach it and subsequent buffered audio cannot overlap it. Explicit saved
+Blob playback continues to use its own AudioContext even when native speech is
+the currently selected provider. `fetchTTS()` and durable `prepareTTS()` report
+`ARCANE_AI_TTS_AUDIO_EXPORT_UNAVAILABLE` for native selection; no audio file is
+fabricated. See [native voice catalogs and previews](ai/browser-speech.md#native-browser-text-to-speech-and-voice-previews).
+
 Playback completion stays pending while the browser waits for an audio-unlock
 gesture or a recoverable resume attempt. If resuming a closed `AudioContext`
 fails, the affected jobs terminate and their playback results settle `false`.
@@ -3731,8 +3749,10 @@ new SpeechPlayback({
 })
 ```
 
-`speech` must expose either `fetchTTS(payload, signal)` or
-`synthesize(payload, {signal})`. `SpeechPlayback` also supplies a third
+`speech` may expose `prepareTTSPlayback(payload,signal)` returning a Blob or
+native descriptor, a native provider's silent `prepare(payload,{signal})`,
+`fetchTTS(payload, signal)`, or `synthesize(payload, {signal})`.
+The playback-oriented method is preferred. `SpeechPlayback` also supplies a third
 SDK-internal preparation object; existing two-argument clients may ignore it.
 `prepare({key,parts,model,voice,responseFormat,
 speed,autoplay=true})` uses only caller-supplied model, voice, and response-format
@@ -3775,6 +3795,15 @@ releases prepared URLs. Replay keeps completed URLs and still-pending provider
 work, then re-submits only failed missing provider segments before starting
 again from index zero.
 
+Native descriptors are retained separately from Blob URLs and never assigned to
+an audio element. `autoplay:false` remains silent. Play, Pause, Resume, Stop and
+Replay use the native control; only a successful current native `finished`
+advances the next part. Native errors remain observable, and a failed native
+control stays owned until its optional `released` promise confirms cleanup.
+Stale completion after stop/restart/disposal cannot advance replacement content.
+Selecting a native part hides and clears the HTML audio source while preserving
+the same shared playback state events and complete original parts.
+
 Every preparation owns an operation ID and one AbortController for each active
 synthesis segment or playback delay. Replacement,
 `stop()`, `cancel()`, and `destroy()` abort their owned signals, suppress stale

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "arcane-os",
-  "version": "0.49.0",
+  "version": "0.50.0",
   "description": "Arcane OS JavaScript SDK, project-local CLI, browser runtime, and repository-portable application packager.",
   "type": "module",
   "main": "./src/index.mjs",

package/runtime/arcane/components/speech.html

@@ -1912,9 +1912,11 @@
             ].includes(error?.code);
     }
 
-    function reportTranscriptionCancellation(reason, message) {
-        const operationId = transcriptionOperationId
-            || nextSpeechOperationId('transcription');
+    function reportTranscriptionCancellation(
+        reason,
+        message,
+        operationId = transcriptionOperationId || nextSpeechOperationId('transcription')
+    ) {
         localStatus = {
             message,
             tone: ''
@@ -1940,6 +1942,10 @@
     }
 
     function cancelSTTOperation(reason) {
+        // A native error can clear the ambient ID before its capture actually ends.
+        const operationId = nativeCaptureSession?.operationId
+            || captureSession?.operationId
+            || transcriptionOperationId;
         const hadActiveOperation = recordingRequested
             || Boolean(captureSession)
             || Boolean(nativeCaptureSession)
@@ -1976,7 +1982,8 @@
                 String(reason || 'stt-transcription-cancelled'),
                 reason === 'component-destroyed'
                     ? 'Transcription stopped because the speech component was destroyed.'
-                    : 'Transcription stopped because the selected service became unavailable.'
+                    : 'Transcription stopped because the selected service became unavailable.',
+                operationId || undefined
             );
         }
         transcriptionOperationId = null;

package/runtime/arcane/modules/AI.js

@@ -1184,6 +1184,7 @@ class AI {
     #ttsSegmentation={...DEFAULT_TTS_SEGMENTATION};
     #preparedSpeechReadiness = null;
     #preparedSpeechPlaybacks = new Set();
+    #nativeSpeechPlaybacks = new Set();
     #speechUnlockContext = null;
 
     #traceSpeech(phase,detail={}){
@@ -5940,6 +5941,11 @@ class AI {
     }
 
     prepareTTS({parts, storage = null, identity = null, signal = null, onState = null} = {}) {
+        if(this.#providerSpeechModel()?.speech?.playback==='native'){
+            const error=new Error('Native browser speech has no audio file to prepare or store; use prepareTTSPlayback.');
+            error.code='ARCANE_AI_TTS_AUDIO_EXPORT_UNAVAILABLE';
+            throw error;
+        }
         if(!is.array(parts)) {
             throw new TypeError('AI.prepareTTS parts must be an array.');
         }
@@ -6460,6 +6466,8 @@ class AI {
             scheduledStart:null,
             sourceNode:null,
             state:'queued',
+            nativePlayback:null,
+            nativeControl:null,
             text,
             voice:options.voice,
             speed:options.speed,
@@ -6530,6 +6538,13 @@ class AI {
             return this.#cancelSpeechJob(job);
         }
 
+        if(audio?.kind==='native-speech'){
+            job.nativePlayback=audio;
+            job.state='ready';
+            this.#requestSpeechPlayback();
+            return true;
+        }
+
         job.state='decoding';
         const audioContext=this.#getSpeechAudioContext();
         return this.playAudio(
@@ -6544,6 +6559,7 @@ class AI {
     async #requestSpeechAudio(job){
         job.abortController??=new AbortController();
         const selection=this.#providerRuntime.selection('tts');
+        const native=this.#providerSpeechModel()?.speech?.playback==='native';
         const voice=job.voice===undefined
             ?(selection?this.#providerSpeechVoice():null)
             :job.voice;
@@ -6551,21 +6567,23 @@ class AI {
             model:selection?.modelId||this.modelTTS,
             input:job.text,
             ...(job.voice!==undefined||voice?{voice}:{}),
-            responseFormat:selection
+            ...(!native?{responseFormat:selection
                 ?this.#providerSpeechResponseFormat()
-                :this.audioFormat,
+                :this.audioFormat}:{}),
             speed:job.speed===undefined?this.voiceSpeed:job.speed
         };
         this.#traceSpeechJob('generation.request',job,{payload,selection});
-        const response=await this.fetchTTS(
+        const response=await this.prepareTTSPlayback(
             payload,
             job.abortController.signal,
             {speechInputPrepared:true}
         );
-        return this.#normalizeProviderSpeechAudio(response);
+        return response?.kind==='native-speech'
+            ?response
+            :this.#normalizeProviderSpeechAudio(response);
     }
 
-    #providerSpeechVoice(){
+    #providerSpeechModel(){
         const selection=this.#providerRuntime.selection('tts');
         if(!selection){
             return null;
@@ -6573,9 +6591,13 @@ class AI {
         const provider=this.#providerRuntime.catalog('tts').find(
             entry=>entry.providerId===selection.providerId
         );
-        const model=provider?.models.find(entry=>entry?.id===selection.modelId);
+        return provider?.models.find(entry=>entry?.id===selection.modelId)??null;
+    }
+
+    #providerSpeechVoice(){
+        const model=this.#providerSpeechModel();
         return is.string(model?.defaultVoice)&&model.defaultVoice.trim()
-            ?model.defaultVoice.trim()
+            ?model.speech?.playback==='native'?model.defaultVoice:model.defaultVoice.trim()
             :null;
     }
 
@@ -6709,7 +6731,38 @@ class AI {
         return this.audioContext;
     }
 
+    async prepareTTSPlayback(payload={},signal=null,preparation={}){
+        const model=this.#providerSpeechModel();
+        if(model?.speech?.playback!=='native'){
+            return this.fetchTTS(payload,signal,preparation);
+        }
+        this.#assertServiceConfigured(this.ttsService,'tts');
+        if(signal?.aborted)throw normalizeAIRequestAbort(signal.reason);
+        const selection=this.#providerRuntime.selection('tts');
+        const request={
+            ...payload,
+            model:payload.model??selection.modelId,
+            voice:payload.voice??this.#providerSpeechVoice(),
+            speed:payload.speed??this.voiceSpeed
+        };
+        const playback=await this.#providerRuntime.request('tts',{
+            operation:'synthesize',payload:request,localOnly:false,signal
+        },{speechInputPrepared:true});
+        if(signal?.aborted)throw normalizeAIRequestAbort(signal.reason);
+        return {
+            ...playback,
+            play(options={}){
+                return playback.play({...options,signal:options.signal??signal});
+            }
+        };
+    }
+
     async fetchTTS(payload={},signal=null,{speechInputPrepared=false}={}){
+        if(this.#providerSpeechModel()?.speech?.playback==='native'){
+            const error=new Error('Native browser speech plays through prepareTTSPlayback and does not export audio.');
+            error.code='ARCANE_AI_TTS_AUDIO_EXPORT_UNAVAILABLE';
+            throw error;
+        }
         const callId=this.#traceSpeech('fetchTTS.call',{payload,aborted:signal?.aborted});
         try{
             this.#assertServiceConfigured(this.ttsService,'tts');
@@ -6958,6 +7011,7 @@ class AI {
             remainder:this.audioMessageChunks,queuedJobs:this.speechJobs.length
         });
         for(const playback of this.#preparedSpeechPlaybacks) playback.stop();
+        for(const playback of this.#nativeSpeechPlaybacks) playback.stop();
         this.speechGeneration+=1;
         this.speechResumeAttempt+=1;
         this.speechResumePending=false;
@@ -6970,6 +7024,7 @@ class AI {
 
         for(const job of this.speechJobs){
             job.abortController?.abort();
+            job.nativeControl?.stop();
             job.state='cancelled';
             this.#traceSpeechJob('queue.cancelled',job,{reason:'stopAudio',callId});
             job.resolvePlayback?.(false);
@@ -6997,7 +7052,7 @@ class AI {
         this.speechJobs.splice(0);
         this.sourceNodes.splice(0);
         this.currentSpeechJob=null;
-        this.isSpeaking=false;
+        this.isSpeaking=this.#nativeSpeechPlaybacks.size>0;
         this.#traceSpeech('stopAudio.result',{callId,result:true});
         return true;
     }
@@ -7011,6 +7066,13 @@ class AI {
             return false;
         }
 
+        if(!audioContext&&this.#providerSpeechModel()?.speech?.playback==='native'){
+            const control=this.currentSpeechJob?.nativeControl;
+            if(control&&control.resume()===false)return false;
+            this.#requestSpeechPlayback();
+            return true;
+        }
+
         if(fromUserGesture){
             this.#clearSpeechUnlock();
         }
@@ -7207,8 +7269,77 @@ class AI {
         );
     }
 
+    async #completeNativeSpeechJob(job,played){
+        if(job.generation!==this.speechGeneration||job.state!=='scheduled')return;
+        if(!played){
+            this.#cancelSpeechJob(job);
+            this.#requestSpeechPlayback();
+            return;
+        }
+        // Playback completion excludes the trailing gap, just as it does for buffers.
+        job.resolvePlayback?.(true);
+        job.resolvePlayback=null;
+        if(job.pauseAfterMs>0){
+            const signal=job.abortController.signal;
+            await new Promise(function waitForNativeSpeechPause(resolve){
+                let timer=null;
+                function finish(){
+                    globalThis.clearTimeout(timer);
+                    signal.removeEventListener('abort',finish);
+                    resolve();
+                }
+                signal.addEventListener('abort',finish,{once:true});
+                if(signal.aborted){finish();return;}
+                timer=globalThis.setTimeout(finish,job.pauseAfterMs);
+            });
+        }
+        if(job.generation!==this.speechGeneration||job.state!=='scheduled')return;
+        this.nextSentance(job);
+    }
+
+    #waitForPrecedingSpeechClock(job){
+        const context=this.speechScheduleContext;
+        const deadline=this.speechScheduleTime;
+        if(!context||deadline<=context.currentTime)return Promise.resolve(true);
+        const signal=job.abortController.signal;
+        return new Promise(function waitForScheduledSpeechGap(resolve,reject){
+            let timer=null;
+            function finish(value,error=null){
+                globalThis.clearTimeout(timer);
+                signal.removeEventListener('abort',cancel);
+                context.removeEventListener('statechange',schedule);
+                if(error)reject(error);
+                else resolve(value);
+            }
+            function cancel(){finish(false);}
+            function schedule(){
+                globalThis.clearTimeout(timer);
+                if(signal.aborted){cancel();return;}
+                const remaining=deadline-context.currentTime;
+                if(remaining<=0){finish(true);return;}
+                if(context.state==='closed'){
+                    finish(false,new Error('The preceding speech audio context closed before its queued pause ended.'));
+                    return;
+                }
+                // A suspended audio clock resumes through its owning state event.
+                if(context.state==='running')timer=globalThis.setTimeout(schedule,remaining*1000);
+            }
+            signal.addEventListener('abort',cancel,{once:true});
+            context.addEventListener('statechange',schedule);
+            schedule();
+        });
+    }
+
+    #releaseNativeSpeechPlayback(control){
+        this.#nativeSpeechPlaybacks.delete(control);
+        this.isSpeaking=this.#nativeSpeechPlaybacks.size>0||this.speechJobs.some(
+            function hasPlayingSpeech(job){return job.state==='scheduled';}
+        );
+        this.#requestSpeechPlayback();
+    }
+
     async #pumpSpeechPlayback(){
-        if(this.speechPlaybackStarting||this.muted){
+        if(this.speechPlaybackStarting||this.muted||this.#nativeSpeechPlaybacks.size){
             return false;
         }
 
@@ -7238,6 +7369,32 @@ class AI {
                     continue;
                 }
 
+                if(job.nativePlayback){
+                    // Native speech has no buffer duration: actual completion owns
+                    // the next turn, including when it follows buffered audio.
+                    if(index!==0||job.state!=='ready')break;
+                    if(!await this.#waitForPrecedingSpeechClock(job)
+                        ||job.generation!==this.speechGeneration||this.muted){
+                        return this.#cancelSpeechJob(job);
+                    }
+                    job.state='scheduled';
+                    this.currentSpeechJob=job;
+                    this.isSpeaking=true;
+                    job.nativeControl=job.nativePlayback.play({signal:job.abortController.signal});
+                    const control=job.nativeControl;
+                    this.#nativeSpeechPlaybacks.add(control);
+                    // A reported native control error need not mean the browser
+                    // released its utterance. Keep Stop ownership until it does.
+                    Promise.resolve(control.released??control.finished).then(
+                        ()=>this.#releaseNativeSpeechPlayback(control),
+                        ()=>this.#releaseNativeSpeechPlayback(control)
+                    );
+                    job.nativeControl.finished.then(
+                        played=>this.#completeNativeSpeechJob(job,played)
+                    ).catch(error=>this.#failSpeechJob(job,error,'playback-start'));
+                    return true;
+                }
+
                 if(!['ready','scheduled'].includes(job.state)||!job.sourceNode?.buffer){
                     break;
                 }
@@ -7333,6 +7490,9 @@ class AI {
             return false;
         }finally{
             this.speechPlaybackStarting=false;
+            const nextJob=this.speechJobs.find(function findFirstUnscheduledSpeechJob(job){
+                return job.state!=='scheduled';
+            });
 
             if(
                 !this.muted
@@ -7347,11 +7507,8 @@ class AI {
                             &&job.scheduledEnd===null;
                     }
                 )
-                &&this.speechJobs.find(
-                    function findFirstUnscheduledSpeechJob(job){
-                        return job.state!=='scheduled';
-                    }
-                )?.state==='ready'
+                &&nextJob?.state==='ready'
+                &&(!nextJob.nativePlayback||this.speechJobs[0]===nextJob)
             ){
                 this.#requestSpeechPlayback();
             }
@@ -7370,7 +7527,7 @@ class AI {
             job.sourceNode.onended=null;
         }
 
-        if(job.scheduledEnd===null&&job.pauseAfterMs>0){
+        if(!job.nativePlayback&&job.scheduledEnd===null&&job.pauseAfterMs>0){
             this.speechScheduleContext=job.audioContext;
             this.speechScheduleTime=
                 (Number(job.audioContext?.currentTime)||0)+job.pauseAfterMs/1000;
@@ -7386,7 +7543,7 @@ class AI {
                 }
             )||null;
         }
-        this.isSpeaking=this.speechJobs.some(
+        this.isSpeaking=this.#nativeSpeechPlaybacks.size>0||this.speechJobs.some(
             function hasScheduledSpeechJob(candidate){
                 return candidate.state==='scheduled';
             }
@@ -7398,9 +7555,14 @@ class AI {
 
     #cancelSpeechJob(job){
         job.abortController?.abort();
+        job.nativeControl?.stop();
         job.state='cancelled';
         this.#traceSpeechJob('queue.cancelled',job);
         this.#removeSpeechJob(job);
+        if(this.currentSpeechJob===job)this.currentSpeechJob=null;
+        this.isSpeaking=this.#nativeSpeechPlaybacks.size>0||this.speechJobs.some(function hasActiveSpeech(candidate){
+            return candidate.state==='scheduled';
+        });
         return false;
     }
 
@@ -7489,7 +7651,7 @@ class AI {
                 }
             )||null;
         }
-        this.isSpeaking=this.speechJobs.some(
+        this.isSpeaking=this.#nativeSpeechPlaybacks.size>0||this.speechJobs.some(
             function hasRemainingScheduledSpeechJob(candidate){
                 return candidate.state==='scheduled';
             }