Teknium bf8004e3a8 fix(stt): kill faster-whisper silence hallucinations at the source ... Local faster-whisper called model.transcribe with bare {'beam_size': 5}: no VAD, cross-window conditioning on, no confidence filtering. Pure silence produced hallucinated tokens (E2E: 5s anullsrc WAV -> 'You', no_speech_prob=0.705) and noisy clips could produce runs of junk, often in other languages. Three-layer class fix, one shared owner for every local-whisper call site (build_local_transcribe_kwargs): 1. Silero VAD filter (bundled with faster-whisper) on by default — silence never reaches the model. stt.local.vad: false restores the raw behavior for music/ambient transcription. stt.local.vad_min_silence_ms tunes chunk splitting (default 500). 2. condition_on_previous_text=False — one hallucinated token can no longer seed a self-reinforcing run; negligible cost for voice-note-length audio. 3. Segment confidence gate (_join_confident_segments): drop a segment only when no_speech_prob > 0.6 AND avg_logprob < -1.0 (openai-whisper's own heuristic shape; both must hit so quiet-but-real speech survives). Config: stt.local.no_speech_prob_threshold / logprob_threshold. The WHISPER_HALLUCINATIONS blocklist in voice_mode.py stays as last-resort defense but should now almost never fire. E2E (real faster-whisper 'base', CPU int8): silence.wav before 'You' -> after '' noise.wav before '' -> after '' speech.wav before/after 'Hello World, this is a test of the transcription system.' (unchanged) Docs (EN + zh-Hans), DEFAULT_CONFIG, cli-config.yaml.example updated; 19 unit tests (kwargs contract, off-switch, confidence gate incl. quiet-speech survival, _transcribe_local wiring), sabotage-verified.		2026-07-29 00:01:33 -07:00
..
computer_use	fix(computer_use): revive ended cua-driver sessions once	2026-07-28 14:24:30 -07:00
environments	fix(environments): exclude multiline session env from terminal snapshots	2026-07-26 19:30:02 -07:00
neutts_samples	…
wakewords	feat(voice): bundle the trained "hey hermes" model as the out-of-the-box default	2026-07-28 07:58:16 -07:00
__init__.py	…
ansi_strip.py	…
approval.py	Reapply "feat(observability): integrate NeMo Relay runtime and shared metrics"	2026-07-27 21:10:51 -07:00
async_delegation.py	feat(delegation): structured stall metadata + live per-child status in /agents	2026-07-26 17:13:52 -07:00
audio_container.py	refactor: extract shared audio container sniffer to tools/audio_container.py	2026-07-28 11:52:44 -07:00
binary_extensions.py	…
blueprints.py	…
browser_camofox.py	…
browser_camofox_state.py	…
browser_cdp_tool.py	fix(browser): stop stale cdp_url from stalling every startup by 10+ seconds	2026-07-27 14:32:05 -07:00
browser_dialog_tool.py	…
browser_supervisor.py	…
browser_tool.py	fix: scope private URL policy per profile	2026-07-28 14:17:45 -07:00
budget_config.py	…
checkpoint_manager.py	feat(diff): cross-surface /diff with staged/all/session modes	2026-07-26 18:28:20 -07:00
clarify_gateway.py	feat(clarify): extend multi-select to gateway text fallback and TUI bridge	2026-07-26 17:46:55 -07:00
clarify_tool.py	fix(clarify): route multi_select through current dispatch paths and harden callback detection	2026-07-26 17:46:55 -07:00
close_terminal_tool.py	…
code_execution_tool.py	fix(kanban): harden delegated-child mutation boundary	2026-07-23 07:33:36 -07:00
computer_use_tool.py	…
credential_files.py	…
cronjob_tools.py	feat(cron): user-owned model pins + cron.model fleet default	2026-07-28 11:52:47 -07:00
daemon_pool.py	…
debug_helpers.py	…
delegate_tool.py	fix(delegate): isolate async batches from parent interrupts	2026-07-28 14:20:10 -07:00
delegation_live_log.py	…
desktop_ui.py	…
discord_tool.py	…
env_passthrough.py	…
env_probe.py	fix(windows): re-fit env_probe console suppression to the temp-file _run + add no-window tests (#67690 follow-up)	2026-07-23 18:20:24 -07:00
fal_common.py	…
feishu_doc_tool.py	…
feishu_drive_tool.py	…
file_operations.py	…
file_state.py	…
file_tools.py	…
focus_pane_tool.py	…
fuzzy_match.py	fix(tools): normalize Unicode space family and minus sign in patch fuzzy matching	2026-07-26 20:58:43 -07:00
homeassistant_tool.py	…
hook_output_spill.py	…
image_generation_tool.py	…
image_source.py	…
interrupt.py	…
kanban_tools.py	fix: scope kanban auto-subscriptions to active profile	2026-07-26 16:27:52 -07:00
lazy_deps.py	fix: route memory-provider dep installs through lazy_deps durable target	2026-07-28 22:40:33 -07:00
managed_tool_gateway.py	fix(windows): sweep remaining bare read_text/write_text sites + linter rule	2026-07-24 17:10:39 -07:00
mcp_dashboard_oauth.py	…
mcp_oauth.py	fix(mcp): make Figma remote OAuth work via DCR allowlist defaults	2026-07-28 00:53:16 -05:00
mcp_oauth_manager.py	fix(mcp): make Figma remote OAuth work via DCR allowlist defaults	2026-07-28 00:53:16 -05:00
mcp_stdio_watchdog.py	…
mcp_tool.py	fix: reject ambiguous MCP tool name collisions	2026-07-28 15:02:52 -07:00
memory_tool.py	fix(memory): close second-read drift race and treat invalid UTF-8 as unreadable	2026-07-24 15:58:01 -07:00
microsoft_graph_auth.py	…
microsoft_graph_client.py	…
neutts_synth.py	fix(voice): reconcile NeuTTS backbone/codec GPU device strings	2026-07-28 14:07:21 -07:00
open_preview_tool.py	…
openrouter_client.py	…
osv_check.py	…
patch_parser.py	…
path_security.py	…
process_registry.py	fix(process_registry): reader loop no longer hangs when an orphaned grandchild holds the stdout pipe	2026-07-24 15:59:02 -07:00
project_tools.py	…
read_extract.py	…
read_terminal_tool.py	…
registry.py	fix: reject ambiguous MCP tool name collisions	2026-07-28 15:02:52 -07:00
schema_sanitizer.py	fix(schema): preserve dependentRequired property names	2026-07-28 14:37:19 -07:00
send_message_tool.py	fix(send_message): pass photon DM chat GUIDs through as explicit targets	2026-07-28 18:21:01 -07:00
session_search_tool.py	fix(session-search): allow scrolling compacted lineage history	2026-07-26 16:18:31 -07:00
skill_manager_tool.py	fix(skills): fail closed on unknown curator ownership	2026-07-25 20:15:37 -07:00
skill_provenance.py	…
skill_usage.py	fix(curator): make the autonomous write policy consistent (#67140)	2026-07-25 19:27:17 -07:00
skills_ast_audit.py	…
skills_guard.py	…
skills_hub.py	refactor: remove the claude-marketplace skill source (redundant Marketplace hub tab)	2026-07-28 23:16:27 -07:00
skills_sync.py	fix(skills): avoid redundant bind-mount scans (#72622)	2026-07-28 04:24:50 -05:00
skills_tool.py	fix(skills): scope 60-char description enforcement to the create path	2026-07-24 07:54:21 -07:00
slash_confirm.py	…
terminal_tool.py	Reapply "feat(observability): integrate NeMo Relay runtime and shared metrics"	2026-07-27 21:10:51 -07:00
thread_context.py	…
threat_patterns.py	…
tirith_security.py	fix: add explicit UTF-8 encoding to all subprocess text=True calls (#53428)	2026-07-24 11:45:57 -07:00
todo_tool.py	…
tool_backend_helpers.py	fix(voice): single scoped resolver — STT/TTS keys fall back to the credential pool	2026-07-28 11:53:09 -07:00
tool_output_limits.py	…
tool_result_storage.py	…
tool_search.py	feat(tool_search): probe-validate blind tool_call args against the deferred schema	2026-07-26 20:59:36 -07:00
transcription_tools.py	fix(stt): kill faster-whisper silence hallucinations at the source	2026-07-29 00:01:33 -07:00
tts_streaming.py	fix(tts): route streaming-provider secrets through resolve_provider_secret; bound per-sentence stream bodies at 16 MiB	2026-07-28 22:31:40 -07:00
tts_text_normalize.py	fix(tts): unify TTS text preprocessing behind one shared cleaner	2026-07-28 11:55:01 -07:00
tts_tool.py	feat(tools): env_passthrough allowlist for command-provider secret scrub	2026-07-28 18:12:26 -07:00
url_safety.py	fix: scope private URL policy per profile	2026-07-28 14:17:45 -07:00
video_generation_tool.py	…
vision_tools.py	fix(security): pin DNS resolutions for SSRF-safe fetches	2026-07-23 11:44:43 -07:00
voice_mode.py	fix(voice): bare stop phrase ends the voice chat on every surface, spoken or typed	2026-07-29 00:01:06 -07:00
wake_word.py	fix(wake): coerce dead onnx->tflite on macOS ARM64; clear stale voice turn-timeout	2026-07-28 19:40:25 -07:00
web_tools.py	fix: add encoding="utf-8" to Path.write_text() calls (P1)	2026-07-24 17:10:39 -07:00
website_policy.py	…
working_diff.py	feat(diff): cross-surface /diff with staged/all/session modes	2026-07-26 18:28:20 -07:00
write_approval.py	…
x_search_tool.py	docs(xai): clarify x_search vs xurl routing without schema cross-refs	2026-07-23 21:06:47 -07:00
xai_http.py	fix(voice): single scoped resolver — STT/TTS keys fall back to the credential pool	2026-07-28 11:53:09 -07:00
xai_video_tools.py	…
yuanbao_tools.py	…