# NousResearch/hermes-agent · Desktop STT transcription timeout too short ## Open now - Nothing open on GitHub. Multiple issues and fixes address the desktop app's audio transcription timeout being too short (defaulting to ~15s), causing local speech-to-text to fail on longer audio. Proposed solutions include overriding, extending, or making the timeout configurable. ## TL;DR - 6 member(s): 0 open issue(s), 2 closed issue(s), 0 open PR(s), 1 merged PR(s), 3 PR(s) closed without merging. - A fix was merged to main in PR #68056. Verify it addresses your case. - **Recommended:** read PR #68056 (and any competing PRs below) before filing. Confirm the merged change covers your symptom. ## Maintainer resolution verdict - Internal-tracker verdict over this cluster's issue members: `unresolved`. This is distinct from the fix status above. Weigh it yourself. ## How this cluster was grouped - Link evidence (`STRUCTURAL_PARTIAL`): some members are connected by structural evidence (shared changed files, or a closing/superseded reference); the rest are joined by an automated relevance judgement over their text. - Members outside that structural core may belong to a different underlying problem. - Pairwise link density `0.29`: the share of member pairs that carry a link, counting a structural link in full and a relevance-only link at 0.36. `1.00` means every pair is linked. ## Closed / duplicate issues - #39448: \[Feature Request\] Configurable timeout for STT transcription (currently hardcoded 15s in Desktop) - #46573: Desktop: transcribeAudio missing timeoutMs — local STT fails for audio >15s ## Prior & competing PRs - **#46579: fix(desktop): override request timeout for transcribeAudio to prevent local STT timeout**. closed without merge · fix tier `suggested` - maintainer direction: category `superseded_by_pr`; implication `superseded`; refs `#46530` - _suggested-tier fix detail withheld; re-request with `?tier=all` to include it._ - **#68056: fix(desktop): extend read aloud + transcription timeouts (salvage of #39286)**. merged to main · fix tier `verified` - maintainer direction: category `merged`; implication `superseded` - files: `apps/desktop/src/hermes.test.ts`, `apps/desktop/src/hermes.ts` - fix hunks: - `apps/desktop/src/hermes.test.ts` ```diff @@ -1,6 +1,12 @@ import { afterEach, beforeEach, describe, expect, it, vi } from 'vitest' import { + AUDIO_SPEAK_MAX_REQUEST_TIMEOUT_MS, + AUDIO_SPEAK_MIN_REQUEST_TIMEOUT_MS, + AUDIO_TRANSCRIBE_MAX_REQUEST_TIMEOUT_MS, + AUDIO_TRANSCRIBE_MIN_REQUEST_TIMEOUT_MS, + audioSpeakRequestTimeoutMs, + audioTranscribeRequestTimeoutMs, getCronJobs, getGlobalModelInfo, getGlobalModelOptions, ``` ```diff @@ -12,7 +18,9 @@ import { listAllProfileSessions, listSessions, listSidebarSessions, - resetSidebarBatchCapability + resetSidebarBatchCapability, + speakText, + transcribeAudio } from './hermes' import { refreshActiveProfile } from './store/profile' ``` ```diff @@ -23,7 +31,7 @@ const emptySessionsResponse = { total: 0 } -describe('Hermes REST session helpers', () => { +describe('Hermes REST helpers', () => { let api: ReturnType beforeEach(() => { ``` ```diff @@ -322,6 +330,62 @@ describe('Hermes REST session helpers', () => { }) }) + it('bounds blocking TTS synthesis timeouts by text length', () => { + expect(audioSpeakRequestTimeoutMs('short message')).toBe(AUDIO_SPEAK_MIN_REQUEST_TIMEOUT_MS) + expect(audioSpeakRequestTimeoutMs('x'.repeat(8_000))).toBe(280_000) + expect(audioSpeakRequestTimeoutMs('x'.repeat(100_000))).toBe(AUDIO_SPEAK_MAX_REQUEST_TIMEOUT_MS) + }) + + it('uses an extended timeout for blocking TTS synthesis', async () => { + api.mockResolvedValueOnce({ + data_url: 'data:audio/mpeg;base64,AA==', + mime_type: 'audio/mpeg', + ok: true, + provider: 'openai' + }) + + await expect(speakText('Read this aloud')).resolves.toEqual({ + data_url: 'data:audio/mpeg;base64,AA==', + mime_type: 'audio/mpeg', + ok: true, + provider: 'openai' + }) + + expect(api).toHaveBeenCalledWith({ + body: { text: 'Read this aloud' }, + method: 'POST', + path: '/api/audio/speak', + timeoutMs: AUDIO_SPEAK_MIN_REQUEST_TIMEOUT_MS + }) + }) + + it('bounds blocking transcription timeouts by payload length', () => { + expect(audioTranscribeRequestTimeoutMs('data:audio/webm;base64,AA==')).toBe(AUDIO_TRANSCRIBE_MIN_REQUEST_TIMEOUT_MS) + expect(audioTranscribeRequestTimeoutMs('x'.repeat(3_000_000))).toBe(300_000) + expect(audioTranscribeRequestTimeoutMs('x'.repeat(9_000_000))).toBe(AUDIO_TRANSCRIBE_MAX_REQUEST_TIMEOUT_MS) + }) + + it('uses an extended timeout for blocking transcription', async () => { + api.mockResolvedValueOnce({ + ok: true, + provider: 'openai', + text: 'transcribed text' + }) + + await expect(transcribeAudio('data:audio/webm;base64,AA==', 'audio/webm')).resolves.toEqual({ + ok: true, + provider: 'openai', + text: 'transcribed text' + }) + + expect(api).toHaveBeenCalledWith({ + body: { data_url: 'data:audio/webm;base64,AA==', mime_type: 'audio/webm' }, + method: 'POST', + path: '/api/audio/transcribe', + timeoutMs: AUDIO_TRANSCRIBE_MIN_REQUEST_TIMEOUT_MS + }) + }) + it('defaults model options to configured providers only', async () => { await getGlobalModelOptions() ``` - `apps/desktop/src/hermes.ts` ```diff @@ -84,6 +84,36 @@ const SESSION_LIST_REQUEST_TIMEOUT_MS = 60_000 // agent-turn ceiling (agent.gateway_timeout = 1800s) so the ack timeout only // ever fires when the turn itself would have been abandoned server-side. export const PROMPT_SUBMIT_REQUEST_TIMEOUT_MS = 1_800_000 +export const AUDIO_SPEAK_MIN_REQUEST_TIMEOUT_MS = 180_000 +export const AUDIO_SPEAK_MAX_REQUEST_TIMEOUT_MS = 600_000 +const AUDIO_SPEAK_TIMEOUT_MS_PER_CHAR = 35 + +export function audioSpeakRequestTimeoutMs(text: string): number { + const estimated = Math.max( + AUDIO_SPEAK_MIN_REQUEST_TIMEOUT_MS, + Math.ceil(String(text || '').length * AUDIO_SPEAK_TIMEOUT_MS_PER_CHAR) + ) + + return Math.min(AUDIO_SPEAK_MAX_REQUEST_TIMEOUT_MS, estimated) +} + +export const AUDIO_TRANSCRIBE_MIN_REQUEST_TIMEOUT_MS = 180_000 +export const AUDIO_TRANSCRIBE_MAX_REQUEST_TIMEOUT_MS = 600_000 +// The transcribe payload is the base64 audio data URL itself, so its string +// length tracks clip size. ~0.1ms/char keeps short clips at the floor while +// letting multi-minute recordings scale toward the cap (a base64 char is +// ~0.75 bytes, so at 128kbps ≈ 21k chars/s of audio this budgets ~2s of +// timeout per 1s of audio before the cap clamps it). +const AUDIO_TRANSCRIBE_TIMEOUT_MS_PER_CHAR = 0.1 + +export function audioTranscribeRequestTimeoutMs(dataUrl: string): number { + const estimated = Math.max( + AUDIO_TRANSCRIBE_MIN_REQUEST_TIMEOUT_MS, + Math.ceil(String(dataUrl || '').length * AUDIO_TRANSCRIBE_TIMEOUT_MS_PER_CHAR) + ) + + return Math.min(AUDIO_TRANSCRIBE_MAX_REQUEST_TIMEOUT_MS, estimated) +} export type { ActionResponse, ``` ```diff @@ -1346,7 +1376,11 @@ export function transcribeAudio(dataUrl: string, mimeType?: string): Promise { return window.hermesDesktop.api({ path: '/api/audio/speak', method: 'POST', - body: { text } + body: { text }, + // TTS blocks until provider synthesis, file read, and base64 encoding + // finish. Remote providers and large messages regularly exceed the + // default 15s Electron backend timeout. + timeoutMs: audioSpeakRequestTimeoutMs(text) }) } ``` - **#46740: fix(desktop): add timeoutMs to transcribeAudio for local STT models**. closed without merge - maintainer direction: category `implemented_elsewhere`; implication `superseded`; refs `2bb531b67a`, `#68056` - **#51862: fix(desktop): extend audio transcription timeout**. closed without merge - maintainer direction: category `duplicate_of`; implication `duplicate`; refs `#46530`, `#46636`, `#46530` ## Fix status - Merged to main in PR #68056 (cited above). No release data is tracked. Verify downstream. - Fix last verified (merged) `2026-07-20`. - Subsystem: `apps/desktop`