arcane-os 0.5.14 → 0.5.16

package/docs/reference/runtime-modules.md

@@ -88,6 +88,7 @@ own asynchronous work, cancellation, and backpressure.
 | [`Mail.js`](#mailjs) | esm | Builds complete reports and prefers the native mail capability with an explicit HTTP transport fallback. | Browser/native hybrid + cloud | Mail inputs/results normalized; transport failures mixed. |
 | [`MailOutbox.mjs`](#mailoutboxmjs) | esm | Persists complete mail reports before delivery and normalizes idempotent enqueue, retry, reconciliation, and invalid-record maintenance. | Browser/native WebView or compatible injected host | Complete records, full work, cancellation, and lifecycle states normalized; storage, lock, and delivery failures coded. |
 | [`MailTransport.mjs`](#mailtransportmjs) | esm | Sends one complete mail report to a normalized HTTP(S) endpoint. | Browser/server with fetch + cloud | Normalized endpoint and transport errors; remote detail preserved. |
+| [`MarkdownSpeech.js`](#markdownspeechjs) | esm | Removes repeated Markdown formatting marks from streamed narration. | Cross-host | Speech-only filtering; single marks and ordinary punctuation remain literal. |
 | [`Marked.min.js`](#markedminjs) | esm | Vendored Marked 18.0.5 Markdown lexer, parser, renderer, extension, and walk-token API. | Cross-host vendor module | Vendor-native Marked contract. |
 | [`MD.js`](#mdjs) | esm | Renders complete Markdown with Marked and exposes the complete rendered markup. | Browser / native WebView | Complete raw and rendered Marked values; parse errors vendor-native. |
 | [`MemoryRecords.js`](#memoryrecordsjs) | esm | Normalizes memory content and detects meaningful stored memory. | Cross-host | Fully normalized string/boolean results. |
@@ -274,7 +275,7 @@ rejected. It propagates the caller-owned signal and returns a playable `Blob`;
 it does not independently choose a provider, cloud fallback, model, runtime, or
 voice policy for the application. `streamTTS(text='',end=false,options={})` and
 `finishTTS()` use this same request boundary. The third-argument options below
-are available in SDK `0.5.12`:
+are available in SDK `0.5.12`, with `textFormat` added in SDK `0.5.16`:
 
 | Field | Default | Meaning |
 | --- | --- | --- |
@@ -282,12 +283,24 @@ are available in SDK `0.5.12`:
 | `speed` | Current `ai.voiceSpeed` | A supplied positive speed is captured for those segments and forwarded to `fetchTTS()`. It does not change `ai.voiceSpeed`. |
 | `pauseAfterMs` | `0` | Finite, nonnegative milliseconds placed after the final extracted segment on the existing audio clock. Invalid values throw `RangeError`; no pause is inserted between this call's other segments. |
 | `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` | `plain` for a new stream | `markdown` removes repeated same formatting marks (`*`, `#`, `_`, backtick, `~`) before segmentation. Single marks and ordinary punctuation remain literal. The selected format lasts through this producer's terminal flush. |
 
 The voice and speed use the existing `fetchTTS()` validation and error path.
-No option rewrites the submitted text. Overrides belong to the segments
+Plain mode preserves the submitted text. Markdown mode changes narration only;
+it leaves displayed, stored, and model content untouched. It is a narrow
+formatting-mark filter, not a full Markdown parser: links, code contents, list
+text, and other characters remain literal. A trailing candidate mark waits for
+the next character so a repeated run split across chunks is still omitted.
+Ordinary prose streams immediately. `end:true`, `finishTTS()`, muted terminal
+calls, and `stopAudio()` clear the formatting state. An explicit format change
+flushes pending formatting marks under the preceding mode before accepting
+the next text. `fetchTTS()` retains its exact-text contract.
+
+Voice, speed, pause, and playback overrides belong to the segments
 extracted in that invocation, including any text buffered by an earlier call.
-Options are not retained with an unfinished `end:false` remainder; a later
-call supplies its own options, and `finishTTS()` uses defaults. Use `end:true`
+Those overrides are not retained with an unfinished `end:false` remainder; a
+later call supplies its own options, and `finishTTS()` uses their defaults
+while retaining the active text format. Use `end:true`
 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`.
@@ -2244,6 +2257,41 @@ import * as module from '/arcane/modules/MailTransport.mjs';
 console.log(Object.keys(module));
 ```
 
+## MarkdownSpeech.js
+
+### Overview
+
+Filters repeated same Markdown formatting marks from narration before speech
+segmentation. The filter recognizes `*`, `#`, `_`, backtick, and `~` runs of two
+or more, including runs arriving across separate chunks. Single marks,
+ellipses, quoted sentence endings, whitespace, and all other text remain
+literal. It neither interprets links nor changes language or voice.
+
+### Public surface
+
+`MarkdownSpeech`; `append(text='',end=false)`, `reset()`.
+
+Exact exports: `MarkdownSpeech`.
+
+### Availability and normalization
+
+**Cross-host.** Plain JavaScript with no dependencies. `append()` returns only
+the newly available narration. Only a trailing candidate marker and whether
+it repeats are retained; ordinary text is emitted immediately. Terminal
+`append(text,true)` flushes a single pending mark and resets state. `reset()`
+clears pending formatting state when its narration is cancelled. Non-string
+input throws `TypeError`.
+
+### Example
+
+```javascript
+import {MarkdownSpeech} from '/arcane/modules/MarkdownSpeech.js';
+
+const speech = new MarkdownSpeech();
+const first = speech.append('**Hello'); // Hello
+const last = speech.append('**...', true); // ...
+```
+
 ## Marked.min.js
 
 ### Overview
@@ -2849,8 +2897,9 @@ console.log(Object.keys(module));
 
 ### Overview
 
-Preserves exact nonblank text as one segment, queues latest-request speech
-synthesis, and controls lookahead HTML audio playback.
+Preserves exact nonblank text, admits complete speech segments according to the
+selected client's advertised capacity, and plays indexed HTML audio in exact
+input order.
 
 ### Public surface
 
@@ -2888,6 +2937,26 @@ freezing it. `prepare()` likewise preserves each nonblank part's exact `input`
 string while normalizing its other playback fields into a new mutable record.
 The class applies no part-count, character-count, pause, or input upper cap.
 
+### Admission and playback order
+
+When `speech` exposes both `fetchTTS(payload, signal)` and
+`providerRuntime.status('tts', {execution:true}).execution.maxConcurrentRequests`
+as a positive safe integer, `prepare()` submits every complete `parts` entry
+immediately. `SpeechPlayback` does not create a second limiter: the provider
+runtime owns bounded FIFO admission. Browser Kokoro defaults that capacity to
+four, so up to four segments can synthesize while later submissions wait in the
+provider queue. A later segment may finish first, but its Blob URL stays at its
+original index and is never played ahead of an earlier segment.
+
+If that capacity is absent, invalid, or unavailable, `SpeechPlayback` retains
+the compatible serialized path and prepares only one lookahead segment. This is
+the default for `Arcane.speech.synthesize` and custom clients, so native hosts
+with one synthesis lock are not driven concurrently. Pause and Resume control
+the same supplied `audio` element. Stop aborts every owned synthesis signal and
+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.
+
 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
@@ -2895,9 +2964,10 @@ settlement, release Blob URLs, and publish synchronous
 `speech-playback-state` occurrences through `globalThis.arcaneEvents`.
 Subscribers receive mutable public state detail. The detail contains
 `state`, `message`, `key`, `index`, `total`, `producing`, `buffered`, `hasAudio`,
-`operationId`, `code`, and `reason`; provider rejection remains preserved to the
-`prepare()` caller. `destroy()` also removes every audio listener and disposes
-its per-instance canonical source handle; repeated destroy returns
+`operationId`, `code`, and `reason`; a first-segment provider rejection remains
+preserved to the `prepare()` caller. Later failures surface when ordered
+playback reaches that segment. `destroy()` also removes every audio listener
+and disposes its per-instance canonical source handle; repeated destroy returns
 `false`. Signal abortion proves delivery suppression; whether provider work
 actually stops remains the selected provider's cancellation boundary.
 
@@ -2934,23 +3004,38 @@ const audio = document.body.appendChild(document.createElement('audio'));
 audio.controls = true;
 const speech = new SpeechPlayback({
   audio,
-  speech: globalThis.ai,
-  model: 'caller-selected-model',
-  voice: 'caller-selected-voice',
-  responseFormat: 'wav'
+  speech: globalThis.ai
 });
-const speakButton = document.body.appendChild(document.createElement('button'));
-speakButton.type = 'button';
-speakButton.textContent = 'Speak';
-speakButton.addEventListener('click', async () => {
+const button = document.body.appendChild(document.createElement('button'));
+button.textContent = 'Speak';
+button.addEventListener('click', async function speakCompleteSegments() {
+  await globalThis.ai.setSpeechMuted(false);
   await speech.prepare({
-    key: 'ready',
-    parts: ['Arcane is ready.'],
+    parts: [
+      'First complete segment.',
+      'Second complete segment.'
+    ],
     autoplay: true
   });
 });
 ```
 
+With the default browser speech configuration, both parts enter its capacity-4
+queue immediately and still play first, then second. Inspect the selected
+execution device without guessing from console warnings:
+
+```javascript
+const execution = globalThis.ai.providerRuntime.status(
+  'tts',
+  { execution: true }
+).execution;
+console.log(execution.selectedDevice, execution.maxConcurrentRequests);
+```
+
+`requestedDevice:'auto'` attempts the complete ONNX Worker/session pool on
+WebGPU and recreates it on WASM if WebGPU loading fails. The reported selected
+device proves provider selection, not physical GPU kernel overlap.
+
 ## StaticDocumentCatalog.js
 
 ### Overview

package/docs/reference/sdk-api.md

@@ -18,7 +18,7 @@ This table is the Node `package.json#exports` map: it defines package
 entrypoints for SDK/tooling code. It is distinct from the generated browser
 import map that resolves application-facing `arcane/*` modules and the focused
 EventManager entry. See [browser runtime delivery](protocols.md#browser-runtime-delivery)
-for the installed-inventory-derived physical-runtime contract in SDK `0.5.14`.
+for the installed-inventory-derived physical-runtime contract in SDK `0.5.16`.
 
 | Specifier | Purpose |
 | --- | --- |
@@ -751,7 +751,7 @@ deterministic map. The package root also contains the public
 {
   schemaVersion: 1,
   kind: 'arcane-app-runtime-projection',
-  sdkVersion: '0.5.14',
+  sdkVersion: '0.5.16',
   pathPrefix: 'arcane/',
   files: [{path}]
 }
@@ -1500,6 +1500,14 @@ specifier `arcane-os/speech-playback` works in Node package consumers and maps
 to the same runtime module in managed browsers. `arcane/SpeechPlayback` is the
 direct browser runtime name generated into the managed import map.
 
+When the supplied `speech` client has `fetchTTS` and reports a positive
+`providerRuntime.status('tts', {execution:true}).execution.maxConcurrentRequests`,
+`prepare()` submits every complete part immediately. The provider owns its
+bounded FIFO queue (browser Kokoro defaults to four), while Blob URLs and
+playback remain in exact input order. A native or custom client without that
+advertised capacity stays serialized with one lookahead. Replay keeps completed
+URLs and pending requests while retrying failed missing provider segments.
+
 ### Signature and result
 
 ```text
@@ -1516,7 +1524,18 @@ runtime module; provider and media availability remain host-owned.
 
 ```javascript
 import SpeechPlayback from 'arcane-os/speech-playback';
-const playback=new SpeechPlayback({audio,speech});
+
+const audio=document.body.appendChild(document.createElement('audio'));
+const playback=new SpeechPlayback({audio,speech:globalThis.ai});
+const button=document.body.appendChild(document.createElement('button'));
+button.textContent='Speak';
+button.addEventListener('click',async function speakCompleteSegments(){
+  await globalThis.ai.setSpeechMuted(false);
+  await playback.prepare({
+    parts:['First complete segment.','Second complete segment.'],
+    autoplay:true
+  });
+});
 ```
 
 ## SPEECH_PLAYBACK_STATE_EVENT
@@ -1546,7 +1565,9 @@ console.log(SPEECH_PLAYBACK_STATE_EVENT);
 
 ### Overview
 
-Named binding for the same canonical class exposed as the speech-playback default. `prepare()` preserves each nonblank part's exact input string without trimming, splitting, or freezing that content.
+Named binding for the same capability-aware, exact-order canonical class exposed
+as the speech-playback default. `prepare()` preserves each nonblank part's exact
+input string without trimming, splitting, or freezing that content.
 
 ### Signature and result
 
@@ -1557,7 +1578,9 @@ new SpeechPlayback(options={})
 ### Availability and normalization
 
 **Node with injected media adapters, or browser/native WebView media.** Binding
-identity equals the default export; provider and media availability remain host-owned.
+identity equals the default export. Provider-advertised capacity enables eager
+submission while that provider owns its bound; native and custom speech stays
+serialized, and media availability remains host-owned.
 
 ### Example
 
@@ -3440,7 +3463,7 @@ workspace it additionally returns the exact installed package authority:
     packageSource,
     canonicalPackageRoot,
     packageName: 'arcane-os',
-    packageVersion: '0.5.14',
+    packageVersion: '0.5.16',
     runtimeRoot,
     browserRuntimeRoot
   }
@@ -3448,9 +3471,9 @@ workspace it additionally returns the exact installed package authority:
 ```
 
 The dependency can be named `arcane-os` or be one exact npm alias for
-`npm:arcane-os@0.5.14`. The selected installation must still be one direct,
+`npm:arcane-os@0.5.16`. The selected installation must still be one direct,
 physical, non-link package directory whose manifest identifies exactly as
-`arcane-os@0.5.14`; duplicate canonical/alias declarations reject.
+`arcane-os@0.5.16`; duplicate canonical/alias declarations reject.
 `allowMissingManagedImportMap` is an internal packaging/development seam. An
 ordinary caller should leave it `false`.
 

package/examples/wasm-ai-demo/README.md

@@ -8,6 +8,8 @@ For the smallest first request, start with the [browser speech quick start](../.
 
 The demo deliberately omits `tts.execution` in `speechConfiguration(dbopfs)` so it consumes the SDK default `{device:'auto',maxConcurrentRequests:4}`. Change that application's TTS record to `execution:{device:'wasm',maxConcurrentRequests:1}` to use less memory, or choose `webgpu` explicitly. Allowed capacities are 1 through 4; Whisper and the LLM retain capacity one.
 
+The maintained Kokoro selection uses `fp32` because [Kokoro.js recommends `fp32` when using WebGPU](https://github.com/hexgrad/kokoro/tree/main/kokoro.js#usage). Automatic fallback keeps the same dtype on WASM. `selectedDevice` identifies the loaded route; it does not validate speech correctness or audio quality, so evaluate actual output for each browser and device combination your application supports.
+
 Capacity 4 means up to four segments synthesize at once. Segment 5 and later wait in the SDK's FIFO queue; they are not dropped. Synthesis may finish out of order, but playback waits for earlier segments and plays exact input order. Each slot owns a Worker/model session, so raising capacity trades memory for latency.
 
 Use the shared Speech component to load/unmute speech, then select **Inspect speech execution** below Chat. The button reads `ai.providerRuntime.status('tts',{execution:true}).execution` and displays requested device, selected device, capacity, active requests, and automatic WASM fallback. It is a snapshot at the moment clicked. `selectedDevice:null` means no pool is selected; `webgpu` reports the upstream execution-provider selection and does not prove physical GPU kernel overlap or audio quality. Auto may fall back to WASM with the same selected model/dtype; explicit WebGPU reports an error when it cannot load.
@@ -39,7 +41,7 @@ Model weights are deliberately excluded from Git. Place the maintained GGUF shar
 
 The maintained model descriptors include each shard's known byte length as progress metadata. During a cold install, the SDK reports aggregate loaded and remaining bytes, transfer speed, estimated time, and active file or Range transfers while Chat presents the existing activation progress bar. Completed shards and completed Range parts within a shard remain available after interruption, so retry downloads only missing work.
 
-The app gives the SDK browser-speech owner the existing Whisper tiny.en FP32 and Kokoro 82M Q8 descriptors. The app does not import those runtimes or create a speech worker itself.
+The app gives the SDK browser-speech owner the existing Whisper tiny.en FP32 and Kokoro 82M FP32 descriptors. The app does not import those runtimes or create a speech worker itself.
 
 ## Local retrieval
 

package/examples/wasm-ai-demo/app.js

@@ -435,12 +435,13 @@ function speechConfiguration(dbopfs) {
     },
     tts: {
       // Omit execution to use the SDK's auto device and four synthesis slots.
+      // Kokoro.js recommends fp32 for the WebGPU route attempted by auto.
       providerId: "wasm-ai-demo-browser-kokoro",
       model: {
         id: "onnx-community/Kokoro-82M-v1.0-ONNX",
         repository: "onnx-community/Kokoro-82M-v1.0-ONNX",
         revision: "1939ad2a8e416c0acfeecc08a694d14ef25f2231",
-        dtype: "q8",
+        dtype: "fp32",
         defaultVoice: "af_heart",
       },
       runtime: {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "arcane-os",
-  "version": "0.5.14",
+  "version": "0.5.16",
   "description": "Arcane OS JavaScript SDK, project-local CLI, browser runtime, and repository-portable application packager.",
   "type": "module",
   "main": "./src/index.mjs",
@@ -64,9 +64,9 @@
     "test": "npm run test:unit && npm run test:functional && npm run test:integration && npm run test:regression",
     "test:release": "node ./bin/arcane-test.mjs test/npm-release.test.mjs",
     "test:unit": "node ./bin/arcane-test.mjs test/app-descriptor.test.mjs test/app-schema.test.mjs test/contracts.test.mjs test/doctor.test.mjs test/mail-credentials.test.mjs test/mail-outbox.test.mjs test/mail-public-api.test.mjs test/mail-send.test.mjs test/mail-transport.test.mjs test/targets.test.mjs test/workspace-operation-lock.test.mjs",
-    "test:functional": "node ./bin/arcane-test.mjs test/browser-speech-providers.test.mjs test/browser-wasm-gpu-notice.test.mjs test/browser-wasm-download-resume.test.mjs test/cli.test.mjs test/dbopfs-document-library.test.mjs test/dev-server.test.mjs test/dom-event-instrumentation.test.mjs test/event-manager.test.mjs test/events.test.mjs test/import-map.test.mjs test/mail-cli.test.mjs test/mail-runtime.test.mjs test/mail-server.test.mjs test/packaging.test.mjs test/persistent-ai-chat-session.test.mjs test/reference-completeness.test.mjs test/runtime-api-behavior.test.mjs test/runtime.test.mjs test/scaffold.test.mjs test/site.test.mjs test/update-check.test.mjs",
+    "test:functional": "node ./bin/arcane-test.mjs test/browser-speech-providers.test.mjs test/browser-wasm-gpu-notice.test.mjs test/browser-wasm-download-resume.test.mjs test/cli.test.mjs test/dbopfs-document-library.test.mjs test/dev-server.test.mjs test/dom-event-instrumentation.test.mjs test/event-manager.test.mjs test/events.test.mjs test/import-map.test.mjs test/mail-cli.test.mjs test/mail-runtime.test.mjs test/mail-server.test.mjs test/packaging.test.mjs test/persistent-ai-chat-session.test.mjs test/reference-completeness.test.mjs test/runtime-api-behavior.test.mjs test/runtime.test.mjs test/scaffold.test.mjs test/speech-playback.test.mjs test/site.test.mjs test/update-check.test.mjs",
     "test:integration": "node ./bin/arcane-test.mjs test/integrated-shared.test.mjs test/integrated-workspace.test.mjs test/mail-browser.test.mjs test/native-plan.test.mjs test/native-provider-loader.test.mjs test/npm-release.test.mjs test/release-bundle.test.mjs test/release-capability-smoke.test.mjs test/shared-payload-batch.test.mjs test/tarball.test.mjs test/wllama-webgpu-runtime.test.mjs",
-    "test:regression": "node ./bin/arcane-test.mjs test/channel-workflows.test.mjs test/logging-regression.test.mjs test/native-provider-generation.test.mjs test/speech-queue-regression.test.mjs test/testing.test.mjs test/test-sets.test.mjs",
+    "test:regression": "node ./bin/arcane-test.mjs test/channel-workflows.test.mjs test/logging-regression.test.mjs test/markdown-speech.test.mjs test/native-provider-generation.test.mjs test/speech-queue-regression.test.mjs test/testing.test.mjs test/test-sets.test.mjs",
     "check": "node tools/check-source.mjs && npm test",
     "check:release": "node tools/check-source.mjs --package-only && npm run test:release",
     "pack:local": "npm pack --ignore-scripts",

package/runtime/arcane/components/chat.html

@@ -2817,7 +2817,9 @@
         ){
             const aiRuntime=boundChatAI??globalThis.ai;
             if(typeof aiRuntime?.streamTTS==='function'){
-                void Promise.resolve(aiRuntime.streamTTS(text)).catch(
+                void Promise.resolve(aiRuntime.streamTTS(
+                    text,false,{textFormat:'markdown'}
+                )).catch(
                     reportTTSError
                 );
             }else{

package/runtime/arcane/modules/AI.js

@@ -13,6 +13,7 @@ import {
 } from './AIProviderRuntime.js';
 import {normalizeOllamaModelIdentifier} from './OllamaModelIdentifier.js';
 import {arcaneLogging} from 'arcane-os/logging';
+import {MarkdownSpeech} from './MarkdownSpeech.js';
 
 const completeValue=(value)=>value;
 
@@ -2029,6 +2030,8 @@ class AI {
     }
 
     audioMessageChunks='';
+    #speechTextFormat='plain';
+    #markdownSpeech=new MarkdownSpeech();
     sourceNodes=[];
     isSpeaking=false;
     audioContext=null;
@@ -5731,6 +5734,8 @@ class AI {
         if(this.muted){
             if(end){
                 this.audioMessageChunks='';
+                this.#speechTextFormat='plain';
+                this.#markdownSpeech.reset();
             }
             this.#traceSpeech('streamTTS.result',{callId,result:false,reason:'muted'});
             return Promise.resolve(false);
@@ -5742,11 +5747,24 @@ class AI {
             throw new RangeError('Speech pauses must be a nonnegative number of milliseconds.');
         }
 
-        this.audioMessageChunks+=String(text||'');
+        const textFormat=options.textFormat??this.#speechTextFormat;
+        if(textFormat!=='plain'&&textFormat!=='markdown'){
+            throw new TypeError('Speech textFormat must be plain or markdown.');
+        }
+        let speechText='';
+        if(textFormat!==this.#speechTextFormat){
+            speechText=this.#markdownSpeech.append('',true);
+        }
+        const incomingText=String(text||'');
+        speechText+=textFormat==='markdown'
+            ?this.#markdownSpeech.append(incomingText,end)
+            :incomingText;
+        this.#speechTextFormat=end?'plain':textFormat;
+        this.audioMessageChunks+=speechText;
         const outputs=this.#extractSpeechSegments(end);
         this.#traceSpeech('streamTTS.segments',{
             callId,segments:outputs,remainder:this.audioMessageChunks,
-            segmentation:this.ttsSegmentation
+            segmentation:this.ttsSegmentation,textFormat,speechText
         });
 
         if(!outputs.length){
@@ -6414,6 +6432,8 @@ class AI {
         this.speechScheduleContext=null;
         this.speechScheduleTime=0;
         this.audioMessageChunks='';
+        this.#speechTextFormat='plain';
+        this.#markdownSpeech.reset();
         this.#clearSpeechUnlock();
 
         for(const job of this.speechJobs){

package/runtime/arcane/modules/MarkdownSpeech.js

@@ -0,0 +1,50 @@
+const FORMATTING_MARKERS=new Set(['*','#','_','`','~']);
+
+// Speech-only filtering; single markers and all other text remain literal.
+class MarkdownSpeech {
+    #pendingMarker='';
+    #repeated=false;
+
+    append(text='',end=false){
+        if(typeof text!=='string'){
+            throw new TypeError('Markdown speech input must be text.');
+        }
+
+        let narration='';
+
+        for(const character of text){
+            if(character===this.#pendingMarker){
+                this.#repeated=true;
+                continue;
+            }
+
+            if(this.#pendingMarker&&!this.#repeated){
+                narration+=this.#pendingMarker;
+            }
+            this.reset();
+
+            if(FORMATTING_MARKERS.has(character)){
+                // Wait for the next character to distinguish one mark from a run.
+                this.#pendingMarker=character;
+            }else{
+                narration+=character;
+            }
+        }
+
+        if(end){
+            if(this.#pendingMarker&&!this.#repeated){
+                narration+=this.#pendingMarker;
+            }
+            this.reset();
+        }
+
+        return narration;
+    }
+
+    reset(){
+        this.#pendingMarker='';
+        this.#repeated=false;
+    }
+}
+
+export {MarkdownSpeech};