Wait for final results from Whisper before completing

<details>
<summary>Prompts Used</summary>
User: This checkout corresponds to https://github.com/jimrandomh/faceclaw/pull/27. Review this branch.

GPT-6 Astra:
> Found one issue:
>
> - **[P2] Wait for Whisper’s final result before completing dictation** — [FaceclawVoiceController.java:729](/Users/jbabcock/repositories/g2/faceclaw/App_Resources/Android/src/main/java/com/faceclaw/app/FaceclawVoiceController.java:729). Skipping partial decodes leaves short utterances without text until release, but the dialog still finalizes after its existing timeout. If decoding exceeds roughly 2.7 seconds (native stop’s 1.5-second wait plus the dialog’s 1.2-second timeout), auto-send can abandon a short command or send an incomplete longer one. Continue can append the late editing instruction literally instead of refining the message. I reproduced these outcomes with delayed transcript events. These flows should await recognition completion.
>
> Validation: TypeScript and a clean Android build passed. All 21 test failures also occur on `main`. Actual Whisper latency was not tested on hardware.

<hr>

User: Thanks, let's fix that issue.

GPT-6 Astra:
> Fixed. Dictation now waits for native recognition to finish before fallback timers run. Auto-send, Continue, and AI Chat preserve delayed Whisper results; cancellation ignores late transcripts.
>
> TypeScript and Android builds pass. All 11 new regression tests pass. The full suite still has the same 21 pre-existing failures.
</details>
This commit is contained in:
James Babcock committed 2026-09-17 22:12:33 -07:00
1 parent 91890a25ac
commit 25c398d60f
11 files changed
+385 -28

No files matched your search

+14 -3
View File
@@ -89,11 +89,22 @@ export class VoiceDraft {
}
private stopAndFinalize(): void {
// Install fallback first: stop may synchronously deliver the final.
this.timer = setTimeout(() => this.finish(), 1500);
const generation = this.generation;
const afterStop = () => {
if (generation !== this.generation || !this.isFinishing()) return;
this.timer = setTimeout(() => this.finish(), 1500);
};
if (this.started) {
this.started = false;
void Promise.resolve(this.actions.stopVoiceCapture()).catch((error) => this.cancel(String(error)));
// Native recognition can take longer than the synchronous stop join.
// Its final callback may finish the draft before this promise resolves.
const stopped = this.actions.stopVoiceCapture();
if (!stopped) { afterStop(); return; }
void stopped.then(afterStop, (error) => {
if (generation === this.generation) this.cancel(String(error));
});
} else {
afterStop();
}
}
+2 -2
View File
@@ -1912,9 +1912,9 @@ class DashboardController {
return this.beginVoiceCapture("ptt", endpointing);
}
private stopVoiceCapture(): void {
private stopVoiceCapture(): Promise<void> | void {
++this.pttCaptureGeneration;
voiceControlBridge.stopPushToTalk();
return voiceControlBridge.stopPushToTalk();
}
private startContinuousVoiceCapture(): void {
+16 -2
View File
@@ -108,6 +108,9 @@ export class FaceclawVoiceControlBridge {
private listening = false;
private detail = "";
private started = false;
private nativeCaptureFinished: Promise<void> = Promise.resolve();
private nextNativeCaptureId = 1;
private readonly nativeCaptureResolvers = new Map<number, () => void>();
// The mic is a single shared stream; these are the reasons it is running.
// The first holder starts capture (choosing the provider); the mic stops
// when the last one releases. Transcript events broadcast to every
@@ -189,8 +192,13 @@ export class FaceclawVoiceControlBridge {
}
/** End push-to-talk: for cloud, commit for a final result if it was the last holder. */
stopPushToTalk(): void {
stopPushToTalk(): Promise<void> | void {
// Shared continuous capture keeps running; cloud providers finalize via
// their own callbacks. Only a stopped native recognizer must be awaited.
const finished = !this.cloudClient && this.captureHolders.has("ptt") && this.captureHolders.size === 1
? this.nativeCaptureFinished : undefined;
this.releaseCapture("ptt", true);
return finished;
}
/** Begin continuous capture (Transcribe): the mic stays on until released. */
@@ -322,7 +330,9 @@ export class FaceclawVoiceControlBridge {
// Which on-device model to load; a no-op setter for every provider except
// "onboard-whisper" (FaceclawVoiceController defaults to Moonshine).
this.controller?.setOnboardModelKind(options.provider === "onboard-whisper" ? "whisper" : "moonshine");
this.controller?.start("onboard");
const captureId = this.nextNativeCaptureId++;
this.nativeCaptureFinished = new Promise((resolve) => { this.nativeCaptureResolvers.set(captureId, resolve); });
this.controller?.start("onboard", captureId);
}
/**
@@ -491,6 +501,10 @@ export class FaceclawVoiceControlBridge {
onTranscript: (text: string, isFinal: boolean) => {
this.emitTranscript(String(text), Boolean(isFinal));
},
onStopped: (captureId: number) => {
this.nativeCaptureResolvers.get(captureId)?.();
this.nativeCaptureResolvers.delete(captureId);
},
onPcm: (pcm: any) => {
const bytes = toUint8Array(pcm);
this.cloudClient?.acceptPcm(bytes);
+1 -1
View File
@@ -28,7 +28,7 @@ export type LayerActions = {
* (hands-free); otherwise it runs until stopVoiceCapture.
*/
startVoiceCapture: (endpointing?: boolean) => Promise<void> | void;
/** Stop push-to-talk; for a cloud provider this also commits for a final result. */
/** Stop push-to-talk; awaits native recognition, or commits a cloud provider for a final result. */
stopVoiceCapture: () => Promise<void> | void;
/** Start continuous capture (Transcribe); shares the mic with push-to-talk. */
startContinuousVoiceCapture: () => Promise<void> | void;
+56 -16
View File
@@ -5,8 +5,7 @@ import { anthropicApiKeySetting } from "../dashboard-settings";
import { GESTURE_CLICK, GESTURE_DOUBLE_CLICK, gestureHints, type InputEvent } from "../gestures";
import { Layer, type LayerActions, type LayerContext } from "../layers";
import { paintInputDialog } from "./input-dialog";
// After the mic stops, the provider's final transcript can trail in (cloud
// commit round-trip); wait this long for it before refining with what we have.
// After native recognition finishes, a cloud final can still trail in.
const FOLLOWUP_FINALIZE_TIMEOUT_MS = 1200;
/**
@@ -80,6 +79,10 @@ export class VoiceInputLayer implements Layer {
private capturing = false;
/** Continuation stopped; waiting for the trailing final transcript. */
private followupFinalizeTimer: ReturnType<typeof setTimeout> | null = null;
private pendingFollowup = false;
private stoppingCapture = false;
private stopGeneration = 0;
private acceptingTranscript = true;
private refineHandle: AnthropicStreamHandle | null = null;
private menuIndex = 0;
/** Auto-send (wakeword skip-confirmation) is waiting to fire. */
@@ -155,21 +158,48 @@ export class VoiceInputLayer implements Layer {
}
this.capturing = false;
this.phase = "menu";
void this.actions.stopVoiceCapture();
if (this.autoSend && this.sendTargets.length) {
// Skip-confirmation (wakeword): send to the default target as soon as the
// transcript finalizes, or after a short wait for the trailing final.
this.pendingAutoSend = true;
this.status = "Sending...";
this.autoSendTimer = setTimeout(() => this.performAutoSend(), FOLLOWUP_FINALIZE_TIMEOUT_MS);
} else if (this.status.endsWith("...")) {
// A progress status ("Starting microphone...", "Listening...") gives
// way to the menu prompt; an error (ending in ".") stays visible.
this.status = "Send, continue, or discard?";
}
this.stopCapture(() => {
if (this.pendingAutoSend) {
this.autoSendTimer = setTimeout(() => this.performAutoSend(), FOLLOWUP_FINALIZE_TIMEOUT_MS);
}
});
this.actions.requestRender();
}
/** Native stop can outlast its synchronous join while Whisper is decoding. */
private stopCapture(onStopped: () => void): void {
const generation = ++this.stopGeneration;
this.stoppingCapture = true;
const afterStop = () => {
if (generation !== this.stopGeneration) return;
this.stoppingCapture = false;
onStopped();
this.actions.requestRender();
};
const stopped = this.actions.stopVoiceCapture();
if (!stopped) {
afterStop();
return;
}
void stopped.then(afterStop, (error) => {
if (generation !== this.stopGeneration) return;
this.stoppingCapture = false;
this.pendingAutoSend = false;
this.acceptingTranscript = false;
this.backToMenu(this.phase === "continuing" ? this.baseText : this.displayText(), String(error));
});
}
/** Fire the queued skip-confirmation send (or fall back to the menu). */
private performAutoSend(): void {
if (this.autoSendTimer !== null) {
@@ -193,7 +223,7 @@ export class VoiceInputLayer implements Layer {
/** The menu rows: one per send target, then Continue, then Discard. */
private menuRows(): Array<{ label: string; dim: boolean; onSelect: () => void }> {
const text = this.displayText().trim();
const hasText = text.length > 0;
const hasText = text.length > 0 && !this.stoppingCapture;
const hasLlmKey = anthropicApiKeySetting.get().trim().length > 0;
const rows: Array<{ label: string; dim: boolean; onSelect: () => void }> = [];
for (const target of this.sendTargets) {
@@ -201,6 +231,7 @@ export class VoiceInputLayer implements Layer {
label: target.label,
dim: !hasText,
onSelect: () => {
if (this.stoppingCapture) return;
this.dismiss();
if (hasText) target.onSend(text);
},
@@ -208,9 +239,9 @@ export class VoiceInputLayer implements Layer {
}
rows.push({
label: hasLlmKey ? "Continue" : "Continue (Needs LLM API key)",
dim: !hasLlmKey,
dim: !hasLlmKey || this.stoppingCapture,
onSelect: () => {
if (hasLlmKey) this.startContinuation();
if (hasLlmKey && !this.stoppingCapture) this.startContinuation();
},
});
rows.push({ label: "Discard", dim: false, onSelect: () => this.dismiss() });
@@ -222,7 +253,7 @@ export class VoiceInputLayer implements Layer {
const inMenu = this.phase === "menu";
paintInputDialog(image, {
title: this.capturing && this.listening ? "Voice ●" : "Voice",
status: this.status,
status: this.stoppingCapture ? "Finishing transcription..." : this.status,
text: this.displayText() || this.placeholderText(),
rows: inMenu ? this.menuRows() : [],
selectedRow: this.menuIndex,
@@ -284,6 +315,7 @@ export class VoiceInputLayer implements Layer {
/** Continue selected: keep the message aside and record a follow-up. */
private startContinuation(): void {
this.acceptingTranscript = true;
this.baseText = this.displayText().trim();
this.finalizedText = "";
this.liveText = "";
@@ -298,15 +330,18 @@ export class VoiceInputLayer implements Layer {
private endContinuationCapture(): void {
if (!this.capturing) return;
this.capturing = false;
void this.actions.stopVoiceCapture();
this.status = "Refining...";
this.pendingFollowup = true;
this.status = "Finishing transcription...";
this.stopCapture(() => {
if (this.pendingFollowup) {
this.followupFinalizeTimer = setTimeout(() => this.beginRefine(), FOLLOWUP_FINALIZE_TIMEOUT_MS);
}
});
this.actions.requestRender();
// The provider's committed transcript arrives shortly after stop; refine
// when it does, or after a timeout with whatever partials we have.
this.followupFinalizeTimer = setTimeout(() => this.beginRefine(), FOLLOWUP_FINALIZE_TIMEOUT_MS);
}
private beginRefine(): void {
this.pendingFollowup = false;
if (this.followupFinalizeTimer !== null) {
clearTimeout(this.followupFinalizeTimer);
this.followupFinalizeTimer = null;
@@ -343,9 +378,10 @@ export class VoiceInputLayer implements Layer {
/** Abort a continuation (mic or LLM stage) and restore the prior message. */
private cancelContinuation(status: string): void {
this.acceptingTranscript = false;
if (this.capturing) {
this.capturing = false;
void this.actions.stopVoiceCapture();
this.stopCapture(() => {});
}
this.refineHandle?.cancel();
this.refineHandle = null;
@@ -353,6 +389,7 @@ export class VoiceInputLayer implements Layer {
}
private backToMenu(text: string, status: string): void {
this.pendingFollowup = false;
if (this.followupFinalizeTimer !== null) {
clearTimeout(this.followupFinalizeTimer);
this.followupFinalizeTimer = null;
@@ -366,6 +403,9 @@ export class VoiceInputLayer implements Layer {
}
onRemoved(): void {
++this.stopGeneration;
this.acceptingTranscript = false;
this.pendingFollowup = false;
this.unsubscribeTranscript?.();
this.unsubscribeTranscript = null;
this.unsubscribeStatus?.();
@@ -428,14 +468,14 @@ export class VoiceInputLayer implements Layer {
private onTranscript(event: VoiceTranscriptEvent): void {
// The refine stream owns the text buffers once it starts; a transcript
// that trails in after that point is stale.
if (this.phase === "refining") return;
if (!this.acceptingTranscript || this.phase === "refining") return;
if (event.isFinal) {
const finalText = event.text.trim() || this.liveText.trim();
if (finalText) {
this.finalizedText = this.finalizedText ? `${this.finalizedText} ${finalText}` : finalText;
}
this.liveText = "";
if (this.phase === "continuing" && this.followupFinalizeTimer !== null) {
if (this.phase === "continuing" && this.pendingFollowup) {
// The follow-up finalized; no need to keep waiting.
this.beginRefine();
return;