@linto-ai/transcript-ui-plugin-live 0.9.23 → 0.9.24-unstable.202610151359

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@linto-ai/transcript-ui-plugin-live",
-  "version": "0.9.23",
+  "version": "0.9.24-unstable.202610151359",
   "description": "Live transcription plugin for @linto-ai/transcript-ui — ingests a real-time partial/final feed",
   "keywords": [
     "vue",
@@ -37,6 +37,6 @@
     "vue": "^3.5.0"
   },
   "dependencies": {
-    "@linto-ai/transcript-ui-core": "0.9.23"
+    "@linto-ai/transcript-ui-core": "0.9.24-unstable.202610151359"
   }
 }

package/src/index.ts

@@ -16,6 +16,11 @@ import type {
 
 const { isSameLanguage, speakText, stopTTS, unlockTTS, isTTSSupported, hasVoices } = utils
 
+/** ASRs can go quiet for a few seconds between the last partial and the final
+ *  while someone is still talking: long enough to bridge that gap, so the
+ *  indicator doesn't blink in the middle of a speech. */
+const DEFAULT_SILENCE_DELAY = 6000
+
 export type { LivePartialEvent, LiveFinalEvent, LiveTranslationEvent }
 export type { LivePluginApi }
 
@@ -53,8 +58,17 @@ function finalEventToTranslationTurn(
   }
 }
 
+export interface LivePluginOptions {
+  /** Whether the voice-playback feature is offered at all. */
+  tts?: boolean
+  /** Milliseconds without a partial before speech counts as over. See
+   *  LivePluginApi.silenceDelay — ASRs differ in how often they emit, so
+   *  hosts tune it per backend. Defaults to 6s. */
+  silenceDelay?: number
+}
+
 export function createLivePlugin(
-  options: { tts?: boolean } = {},
+  options: LivePluginOptions = {},
 ): CorePlugin {
   // Deployment flag: whether the voice-playback feature is offered at all.
   const ttsAvailable = options.tts ?? false
@@ -64,6 +78,10 @@ export function createLivePlugin(
 
     install(core: Core) {
       const partial = shallowRef<string | null>(null)
+      const partialsVisible = ref(true)
+      const isSpeechActive = ref(false)
+      const silenceDelay = ref(options.silenceDelay ?? DEFAULT_SILENCE_DELAY)
+      let silenceTimeout: ReturnType<typeof setTimeout> | null = null
       const hasLiveUpdate = ref(false)
       // Voice playback of finalized turns (browser speech synthesis).
       const ttsEnabled = ref(false)
@@ -84,12 +102,49 @@ export function createLivePlugin(
 
       hasLiveUpdate.value = true
 
+      // Partials are the only sign that someone is talking, and nothing on
+      // the wire ever announces silence — it is only the absence of the next
+      // one. Each partial pushes the deadline back, so the state doesn't
+      // flicker off between two sentences.
+      function markSpeechActivity(): void {
+        isSpeechActive.value = true
+        if (silenceTimeout !== null) clearTimeout(silenceTimeout)
+        silenceTimeout = setTimeout(endSpeechActivity, silenceDelay.value)
+      }
+
+      function endSpeechActivity(): void {
+        if (silenceTimeout !== null) {
+          clearTimeout(silenceTimeout)
+          silenceTimeout = null
+        }
+        isSpeechActive.value = false
+      }
+
       function clearPartial(): void {
         // shallowRef detects the change on its own, no triggerRef needed
         partial.value = null
         lastOriginalPartialEvent = null
       }
 
+      // The single door every partial goes through — source, translation and
+      // cross alike — so one setting closes it for the transcription panel
+      // and the subtitle scroller at once. Ignored while they are hidden.
+      function setPartial(text: string): void {
+        if (!partialsVisible.value) return
+        partial.value = text
+      }
+
+      function showPartials(): void {
+        partialsVisible.value = true
+      }
+
+      function hidePartials(): void {
+        partialsVisible.value = false
+        // Drop what is already on screen now, rather than leaving the last
+        // provisional line up until a final turn happens to replace it.
+        immediateClearPartial()
+      }
+
       function isTranslationTrackFor(
         active: TranslationInfo,
         language: string,
@@ -103,6 +158,10 @@ export function createLivePlugin(
         const channel = core.activeChannel.value
         if (!channel) return
 
+        // Someone is talking on the channel being watched — true whichever
+        // track is displayed, and whether or not the text itself is shown.
+        markSpeechActivity()
+
         lastOriginalPartialEvent = event
 
         const activeTranslation = channel.activeTranslation.value
@@ -110,7 +169,7 @@ export function createLivePlugin(
         // Only the original (source-language) partial flows through here.
         // Translated partials arrive via onTranslation, one per translation.
         if (activeTranslation.isSource && event.text != null) {
-          partial.value = event.text
+          setPartial(event.text)
         }
       }
 
@@ -247,6 +306,8 @@ export function createLivePlugin(
         const activeTranslation = channel.activeTranslation.value
 
         if (!_event.final) {
+          markSpeechActivity()
+
           if (activeTranslation.id === CROSS_TRANSLATION_ID) {
             if (
               _event.turnId === lastOriginalPartialEvent?.turnId &&
@@ -255,10 +316,10 @@ export function createLivePlugin(
                 lastOriginalPartialEvent?.language,
               )
             ) {
-              partial.value = _event.text
+              setPartial(_event.text)
             }
           } else if (isTranslationTrackFor(activeTranslation, _event.language)) {
-            partial.value = _event.text
+            setPartial(_event.text)
           }
           return
         }
@@ -305,6 +366,11 @@ export function createLivePlugin(
 
       const api: LivePluginApi = {
         partial,
+        partialsVisible,
+        isSpeechActive,
+        silenceDelay,
+        showPartials,
+        hidePartials,
         hasLiveUpdate,
         ttsAvailable,
         ttsEnabled,
@@ -336,6 +402,7 @@ export function createLivePlugin(
 
       return () => {
         immediateClearPartial()
+        endSpeechActivity()
         stopTTS()
         if (ttsSupported) {
           window.speechSynthesis.removeEventListener(

package/src/test/createLivePlugin.test.ts

@@ -0,0 +1,164 @@
+import { describe, expect, it } from "bun:test"
+import { createCore } from "@linto-ai/transcript-ui-core"
+import type { Core } from "@linto-ai/transcript-ui-core"
+import { createLivePlugin } from "../index"
+
+/** One channel, a French source and an English translation track — enough
+ *  for both partial paths (source, and translated/cross). */
+function makeLiveCore(silenceDelay?: number): Core {
+  const core = createCore()
+  core.setDocument({
+    title: "live",
+    speakers: new Map(),
+    channels: [
+      {
+        id: "ch-1",
+        name: "channel 1",
+        duration: 0,
+        translations: [
+          { id: "tr-fr", languages: ["fr"], isSource: true, turns: [] },
+          { id: "tr-en", languages: ["en"], isSource: false, turns: [] },
+        ],
+      },
+    ],
+  })
+  core.use(createLivePlugin(silenceDelay ? { silenceDelay } : {}))
+  return core
+}
+
+function wait(ms: number): Promise<void> {
+  return new Promise((resolve) => setTimeout(resolve, ms))
+}
+
+function sendPartial(core: Core, text: string): void {
+  core.live!.onPartial({ turnId: "turn-1", text, language: "fr" }, "ch-1")
+}
+
+function sendTranslatedPartial(core: Core, text: string): void {
+  core.live!.onTranslation({
+    turnId: "turn-1",
+    language: "en",
+    sourceLanguage: "fr",
+    text,
+    final: false,
+    startTime: 0,
+    endTime: 1,
+    speakerId: null,
+  })
+}
+
+describe("createLivePlugin — partials visibility", () => {
+  it("shows partials by default", () => {
+    const core = makeLiveCore()
+
+    expect(core.live!.partialsVisible.value).toBe(true)
+
+    sendPartial(core, "bonjour le")
+    expect(core.live!.partial.value).toBe("bonjour le")
+  })
+
+  it("drops what is on screen and swallows the next ones once hidden", () => {
+    const core = makeLiveCore()
+    sendPartial(core, "bonjour le")
+
+    core.live!.hidePartials()
+    expect(core.live!.partialsVisible.value).toBe(false)
+    expect(core.live!.partial.value).toBeNull()
+
+    sendPartial(core, "bonjour le monde")
+    expect(core.live!.partial.value).toBeNull()
+  })
+
+  it("lets them through again when shown back", () => {
+    const core = makeLiveCore()
+    core.live!.hidePartials()
+    sendPartial(core, "swallowed")
+
+    core.live!.showPartials()
+    expect(core.live!.partialsVisible.value).toBe(true)
+
+    sendPartial(core, "bonjour")
+    expect(core.live!.partial.value).toBe("bonjour")
+  })
+
+  it("gates translated partials too, not just the source ones", () => {
+    const core = makeLiveCore()
+    core.activeChannel.value!.setActiveTranslation("tr-en")
+
+    sendTranslatedPartial(core, "hello the")
+    expect(core.live!.partial.value).toBe("hello the")
+
+    core.live!.hidePartials()
+    expect(core.live!.partial.value).toBeNull()
+
+    sendTranslatedPartial(core, "hello the world")
+    expect(core.live!.partial.value).toBeNull()
+  })
+
+  it("still records finalized turns while partials are hidden", () => {
+    const core = makeLiveCore()
+    core.live!.hidePartials()
+
+    core.live!.onFinal(
+      {
+        turnId: "turn-1",
+        speakerId: null,
+        text: "bonjour le monde",
+        words: [],
+        startTime: 0,
+        endTime: 2,
+        language: "fr",
+      },
+      "ch-1",
+    )
+
+    const turns = core.activeChannel.value!.sourceTranslation.turns.value
+    expect(turns.map((t) => t.text)).toEqual(["bonjour le monde"])
+    expect(core.live!.partial.value).toBeNull()
+  })
+})
+
+describe("createLivePlugin — speech activity", () => {
+  it("lights up on a partial and goes out after the silence delay", async () => {
+    const core = makeLiveCore(30)
+
+    expect(core.live!.isSpeechActive.value).toBe(false)
+
+    sendPartial(core, "bonjour")
+    expect(core.live!.isSpeechActive.value).toBe(true)
+
+    await wait(15)
+    // Still talking: each partial pushes the deadline back.
+    sendPartial(core, "bonjour le")
+    await wait(20)
+    expect(core.live!.isSpeechActive.value).toBe(true)
+
+    await wait(40)
+    expect(core.live!.isSpeechActive.value).toBe(false)
+  })
+
+  it("tracks speech even while the partial text is hidden", async () => {
+    const core = makeLiveCore(30)
+    core.live!.hidePartials()
+
+    sendPartial(core, "bonjour")
+    expect(core.live!.partial.value).toBeNull()
+    expect(core.live!.isSpeechActive.value).toBe(true)
+
+    await wait(50)
+    expect(core.live!.isSpeechActive.value).toBe(false)
+  })
+
+  it("counts translated partials as speech too", () => {
+    const core = makeLiveCore(30)
+    core.activeChannel.value!.setActiveTranslation("tr-en")
+
+    sendTranslatedPartial(core, "hello")
+    expect(core.live!.isSpeechActive.value).toBe(true)
+  })
+
+  it("defaults to a two-second delay, host-tunable at runtime", () => {
+    expect(makeLiveCore().live!.silenceDelay.value).toBe(6000)
+    expect(makeLiveCore(500).live!.silenceDelay.value).toBe(500)
+  })
+})