BuzzbeeSCD/openchamber
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat(dictation): transcribe after recording instead of live

Browse Source 
Parakeet is an offline model trained on whole utterances, so re-decoding
the growing buffer to animate a live transcript cost O(n^2) work for a
result the final decode replaced. Sessions now decode once per committed
segment, and the composer shows a scrolling waveform of the mic level
instead of running text.

Long dictations split at a pause once past 60s (hard cap 90s) instead of
on a blind 15s timer, so cuts no longer land mid-word. Committed segments
decode while the user is still speaking: a 185s dictation returns 4.1s
after stop instead of 11.0s, with identical text (816 vs 817 words).

Also fixes two ways the stream manager could silently drop transcribed
audio. It now counts the commits it issued instead of trusting the
session's echoed events, so a commit still in flight when the client
finishes can no longer be left out of the final text. And segment
byte/peak accounting is reset where the commit is issued rather than when
the event arrives, which could mistake the tail of a dictation for
silence and clear it.
This commit is contained in:
Bohdan Triapitsyn
2026-08-22 01:10:19 +03:00
parent 35998f9f4d
commit 23928d342c
11 changed files with 393 additions and 168 deletions
Download Patch File Download Diff File Expand all files Collapse all files
packages/ui/src/hooks/useDictation.ts
+15 -8
View File
@@ -2,16 +2,19 @@
* Streaming dictation state machine.
*
* Status flow: idle -> recording -> uploading -> idle | failed.
* While recording, mic PCM chunks stream to the server, which sends back live
* partial transcripts. Confirm finalizes and resolves the full text; failed
* dictations retain their audio segments so retry can replay them.
* While recording, mic PCM chunks stream to the server, which transcribes them
* segment by segment; confirm finalizes and resolves the full text. Nothing is
* shown while recording — `partialTranscript` holds whatever the server has
* transcribed so far and exists only so a failed dictation can be salvaged
* instead of losing minutes of speech. Failed dictations also retain their
* audio segments so retry can replay them.
*/
import { useCallback, useEffect, useRef, useState } from 'react';
import { dictationClient, type DictationStartOptions } from '@/lib/dictation/dictation-client';
import { DictationStreamSender } from '@/lib/dictation/dictation-stream-sender';
import { useDictationAudioSource } from '@/lib/dictation/use-dictation-audio-source';
import { useDictationAudioSource, type DictationLevelListener } from '@/lib/dictation/use-dictation-audio-source';
import { useConfigStore } from '@/stores/useConfigStore';
export type DictationStatus = 'idle' | 'recording' | 'uploading' | 'failed';
@@ -26,8 +29,10 @@ export interface UseDictationResult {
status: DictationStatus;
isRecording: boolean;
isProcessing: boolean;
/** Server-side transcript so far; recovery only, never shown while recording. */
partialTranscript: string;
volume: number;
/** Subscribe to the normalized (0..1) mic level for the waveform. */
subscribeLevel: (listener: DictationLevelListener) => () => void;
duration: number;
error: string | null;
errorReason: string | null;
@@ -133,7 +138,8 @@ export function useDictation(options: UseDictationOptions = {}): UseDictationRes
setPartialTranscript('');
}, []);
// Live partial transcripts for the active dictation.
// Transcripts of segments the server has already committed. Not rendered
// while recording; kept so a failed dictation can be salvaged.
useEffect(() => {
return dictationClient.onPartial((dictationId, text) => {
const activeDictationId = senderRef.current?.getDictationId();
@@ -223,7 +229,8 @@ export function useDictation(options: UseDictationOptions = {}): UseDictationRes
try {
await audio.start();
startDurationTracking();
// Open the stream eagerly so partials start flowing immediately.
// Open the stream eagerly so audio uploads while the user speaks
// and only the tail is left to transcribe on stop.
await senderRef.current?.restartStream().catch((err) => {
// Non-fatal: segments buffer locally and finish() retries the
// start, but surface the reason (e.g. model downloading) so
@@ -394,7 +401,7 @@ export function useDictation(options: UseDictationOptions = {}): UseDictationRes
isRecording: status === 'recording',
isProcessing: status === 'uploading',
partialTranscript,
volume: audio.volume,
subscribeLevel: audio.subscribeLevel,
duration,
error,
errorReason,