* feat: add remote API speech recognition engine with installation and configuration support.
* fix: reformatted by black
* fix: unit test in test_main_initializes_components
* fix typo
* Apply suggestion from @jatinkrmalik
Co-authored-by: Jatin K Malik <jatinkrmalik@gmail.com>
* feat: Validate remote API URL format, enhance connection testing, update default remote API endpoint.
* fix typo
* feat: add remote API recognition manager test suite
* fix: add Python 3.14 support and lxml>=6.1.0 (fixes#404)
- Add Python 3.14 to classifiers
- Add lxml>=6.1.0 to dependencies to fix install on Python 3.14 (CachyOS, Arch)
- Update black target-version to include py314
Fixes#404Closes#405
* fix(install): validate pyproject.toml/setup.py content before entering local repo mode (fixes#396)
Fix false local repo detection when users run `uv init` before installation. Changed detection from naive file existence check to content validation:
- For pyproject.toml: check for `name = "vocalinux"`
- For setup.py: check for "vocalinux" string
- Falls through to remote clone mode if neither matches
Fixes#396
* fix(recognition): preserve final speech on stop (fixes#401)
Replace hard-coded stop tail trim with configurable stop_sound_guard_ms
(default 200ms) so quick key release no longer clips the final spoken
word in push-to-talk mode.
Co-authored-by: Fakhareddine <fe.sabbari@claranet.com>
* fix(ibus): handle engine instance destruction on layout switch (fixes#388, closes#389)
Override do_destroy() to clear _active_instance when IBus destroys the
engine on keyboard layout switch (Super+Space). Add retry logic in
inject_text() — on NO_ENGINE, re-activates via switch_engine() and
retries once to recover cleanly.
Co-authored-by: Sbenazar <127324689+Sbenazar@users.noreply.github.com>
* chore: ignore opentree worktree dir and config (#412)
.opentree/ contains git worktrees created by the opentree CLI tool
(https://github.com/axelgar/opentree). These are local developer
artifacts and should not be tracked.
opentree.toml is the per-repo opentree config - also local/personal,
not something contributors need to share.
* test(ibus): fix do_destroy CI failure (issubclass TypeError without IBus) (#410)
* test(ibus): fix do_destroy super() test failing in CI without IBus installed
Patch builtins.super instead of relying on mock_ibus.Engine.do_destroy,
which fails with issubclass() TypeError when VocalinuxEngine is built
with object as its base (IBus unavailable in CI).
* test(ibus): fix MagicMock.__new__ TypeError on Python < 3.13
VocalinuxEngine inherits from MagicMock at import time (IBus is mocked
via sys.modules at test-file level). Calling VocalinuxEngine.__new__(VocalinuxEngine)
delegates to MagicMock.__new__ which runs issubclass(cls, AsyncMockMixin).
On Python < 3.13 that raises TypeError when the module-cached class is a
MagicMock spec object rather than a real type.
Fix: introduce _make_engine() helper that allocates via a thin concrete
subclass whose __new__ calls object.__new__ directly, bypassing the
MagicMock allocation path entirely. Verified across all three
TestVocalinuxEngineDestroy test methods.
* test(ibus): use unbound do_destroy call to avoid MagicMock interference
The previous fix (thin _ConcreteEngine subclass) still failed because
VocalinuxEngine inherits MagicMock.__getattribute__, which intercepts
.do_destroy on instances and returns a mock attribute rather than the
real method - regardless of Python version.
Call VocalinuxEngine.do_destroy(engine) directly with a plain object()
as the self argument instead. do_destroy only uses self for identity
comparison (is VocalinuxEngine._active_instance), so any object works.
This bypasses MagicMock's MRO entirely.
* test(ibus): reload module and instantiate real engine class for do_destroy tests
* refactor(ibus): extract destroy handler for stable cross-version tests
* test(ibus): cover destroy helper branches and do_destroy delegation
* test(ibus): stabilize do_destroy CI behavior across environments
Avoid brittle super() mocking in no-IBus environments and gate the IBus-only parent destroy path test behind runtime capability checks to prevent false CI failures on Python matrix runners.
* test(ibus): remove flaky do_destroy class-path assertions
Drop brittle class-method destroy tests that fail under mocked GI/IBus on older Python runners and keep deterministic helper-level coverage for destroy behavior.
* fix(text-injection): recover from IBus runtime failures (#411)
* fix(ibus): retry and recover from transient socket failures
Harden IBus injection against connection-refused, timeout, and engine process races so dictation is more resilient in real-world desktop sessions.
Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent)
* fix(text-injection): fallback to non-IBus backend on runtime IBus failures
Allow dictation to continue when IBus runtime injection fails by automatically switching to xdotool/wtype/ydotool fallbacks instead of returning an immediate failure.
Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent)
* fix(ibus): preserve live socket injection during recovery
Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent)
* test(coverage): add runtime fallback and ibus retry coverage
Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent)
* fix(ibus): probe engine readiness at startup and harden NO_ENGINE retries (#391)
* fix(ibus): wait for engine readiness before first injection
Add a startup readiness probe that pings the IBus engine after activation so first-use dictation no longer races do_enable(). Also switch NO_ENGINE retry behavior to exponential backoff and cover the new flow in tests.
Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent)
Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>
* test(ibus): cover readiness probe edge cases
Add focused tests for the new startup readiness path, including direct PING success, retry exhaustion, unexpected responses, and engine re-activation before injection. This raises coverage on vocalinux.text_injection.ibus_engine from 68% to 73% in the targeted suite.
Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent)
Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>
* fix(text-injection): avoid blocking IBus and clipboard fallbacks
Warm IBus in the background so startup no longer stalls while readiness probing, and make clipboard helpers fail fast with per-session health caching so non-ASCII ydotool fallback does not lock dictation.
Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent)
Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>
* test(ibus): update retry tests for merged NO_ENGINE recovery logic
---------
Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>
* chore(deps-dev): bump postcss from 8.4.39 to 8.5.10 in /web
Bump postcss dev dependency (CSS processing). No runtime impact.
* fix(logging): clean up runtime log noise and cache hardware detection
Clean up duplicated code artifacts from recent merges and cache hardware detection to eliminate ~1.3s UI lag when opening settings dialog.
* test: cover recognition internals and suppress desktop notifications (#414)
* test(recognition): cover _perform_recognition and _enqueue_audio_segment
The previous commit demoted logger.info('DEBUG: ...') to logger.debug(...)
and removed a dead loop body in _perform_recognition. Codecov flagged 12
uncovered lines — all inside _perform_recognition and _enqueue_audio_segment,
which existing tests always mocked out.
Add direct (non-mocked) tests for both methods:
- _enqueue_audio_segment:
* empty buffer → early return
* normal buffer → enqueues correctly
* full queue → drops oldest and retries
- _perform_recognition:
* processes a segment then exits on None signal
* drains remaining items on None signal
* exits cleanly when idle with empty queue
* continues looping on queue timeout while recording
These tests exercise the actual threading/queue logic and cover the
logger.debug / logger.warning calls as a side effect.
* test(conftest): suppress real desktop notifications during test runs
_show_notification shells out to notify-send, which pops up real system
notifications when tests exercise the model-not-ready path. Add an autouse
fixture that patches _show_notification for all tests EXCEPT those whose
name contains 'show_notification' (so direct tests of the helper still
verify subprocess.Popen calls).
* chore(gitignore): ignore .sisyphus directory
* fix(install): improve openSUSE dependency handling with fallbacks (#418)
The suse case in install_system_dependencies() previously installed all
packages blindly, causing the entire install to fail if a single package
was unavailable (e.g., glslang on some openSUSE repos).
Changes:
- Check each package individually with rpm -q before installing
- Add graceful fallback for glslang -> glslang-devel
- Add graceful fallback for libappindicator-gtk3 -> typelib-1_0-AppIndicator3-0_1 -> libayatana-appindicator3-1
- Continue installation with warning instead of hard failure
Fixes#407
* fix(remote_api): re-init engine on resume from suspend
Add a remote_api branch to reinitialize_after_resume mirroring the
dispatch chain in __init__, so the engine is recreated after a
suspend/resume cycle. Null self._http_session alongside the existing
self.model and self.recognizer resets so a stale pooled-connection
session is not reused after resume.
* perf(remote_api): pool HTTP connections via requests.Session
Introduces a manager-owned requests.Session (_http_session) so all
remote_api engine HTTP traffic reuses a single connection pool instead
of opening a new connection for every request.
Lifecycle:
- __init__: self._http_session = None (attribute always present)
- _init_remote_api: close prior session if it exists, then create a
new requests.Session(); use session.get for the connection test.
Prevents connection leaks on back-to-back inits (re-init guard).
- _try_openai_api / _try_whispercpp_server_api: replaced requests.post
with self._http_session.post; local 'import requests' retained for
the requests.exceptions.ConnectionError catch.
- reconfigure: captures old engine before self.engine is updated;
closes and nulls _http_session when switching away from remote_api.
- reinitialize_after_resume: nulls _http_session before _init_remote_api
recreates it.
Tests: added TestHTTPSession class (4 cases) in
tests/test_recognition_manager_remote_api.py; all 40 tests pass.
Updated _setup_requests_get_*/post_* helpers to configure the session
instance (mock_requests.Session.return_value) instead of the
module-level mock, keeping all 36 pre-existing tests green.
* docs: add remote HTTP transcription setup guide
* fix(install): avoid asking for remote server URL twice
run_interactive_install collects the URL upfront in Step 3, but
install_python_package re-declared REMOTE_API_URL as local and prompted
again, ignoring the wizard value. Inherit the wizard's value instead.
* ui(settings): hide remote API behind an Advanced tab
Addresses upstream PR #335 review feedback from @jatinkrmalik that
remote API is not plug-and-play and should not be a peer of the local
engines for casual users.
- Remove 'Remote API' from the Speech Engine tab's engine combo; the
combo now shows local engines only (vosk / whisper / whisper_cpp).
- Add a new 'Advanced' notebook tab with a 'Remote Server' group
containing a 'Use remote server' switch, the existing URL/Key/
Endpoint fields, and the Test Connection button + status label.
- The switch overrides the local engine selection. get_selected_settings
returns engine='remote_api' iff the switch is on, otherwise the combo
selection. URL/Key/Endpoint persist independently of the toggle state.
- _load_and_apply_settings: if config has engine='remote_api', the
switch is turned on and the combo is set to a sensible local fallback
(whisper_cpp, or first available local engine) so toggling off has
somewhere to land.
- _update_engine_specific_ui no longer special-cases the combo for
remote_api (the combo never has it); it hides the model row, info
card and legend whenever the remote switch is on.
- Update HTTP_REMOTE.md to reflect the new tab location.
* ui(remote-api): mark remote API as advanced and default toggle off
- install.sh: relabel option 4 as 'REMOTE API (ADVANCED)' to signal
it as a non-default, advanced engine choice.
- settings_dialog.py: explicitly initialise the 'Use remote server'
switch to inactive at widget creation, making the default state
unambiguous (saved-config load logic still toggles it on for
existing remote_api installs).
* fix(remote-api): preserve trailing newline from transcription responses
The remote-API transcription path stripped all whitespace from the
upstream response before injection, silently discarding a trailing
newline that some pipelines (post-processing proxies, fixup layers)
emit to signal "submit" / Enter. Whisper itself doesn't emit trailing
newlines, so for the common case this is a no-op; for users with a
post-processing layer it stops their submit signal from being eaten.
_filter_non_speech now uses a whitespace-stripped probe for its
non-speech pattern checks but returns probe + "\n" when the original
input ended with "\n" or "\r" (CRLF normalises to LF). The redundant
.strip() at the call site is removed so the function receives the raw
API text. Existing filter behaviour (BLANK_AUDIO, music notes, low
speech-content ratio, etc.) is unchanged. Four new unit tests cover
the trailing-newline cases.
* test(remote-api): cover transcribe exception handler and unknown-engine reinit branch
Adds two unit tests buttoning up previously-uncovered defensive branches:
- _transcribe_with_remote_api swallows unexpected exceptions and returns
"" — verified by raising RuntimeError from _try_whispercpp_server_api.
- reinitialize_after_resume short-circuits without dispatching when the
engine attribute is unknown — verified by mutating engine to a bogus
value and asserting no _init_* method is called.
* feat(settings): add Advanced tab with whisper.cpp anti-hallucination parameters (#415)
* feat(settings): add Advanced tab with whisper.cpp anti-hallucination parameters
Add a new Advanced settings tab that exposes whisper.cpp decoding
parameters to help users reduce hallucinations and tune recognition
behavior.
Parameters exposed:
- No Timestamps (default: on)
- Suppress Non-Speech Tokens (default: on)
- No Context (default: on)
- Initial Prompt (default: empty)
- Temperature (default: 0.0)
- Temperature Increment (default: -1.0, disables fallback)
- Entropy Threshold (default: 2.4)
- Logprob Threshold (default: -1.0)
- No-Speech Threshold (default: 0.6)
All parameters are saved under the advanced config section and are
only active when the whisper.cpp engine is selected. The UI grays out
these settings and shows an explanatory note when another engine is
active.
Changes:
- config_manager.py: add whisper.cpp keys to DEFAULT_CONFIG[advanced]
- recognition_manager.py: read params from kwargs/config, pass to Model()
- main.py: pass advanced settings from config to SpeechRecognitionManager
- settings_dialog.py: new Advanced tab with switches, spin buttons, entry
- tests: update test_main.py assertions, add whispercpp_advanced_defaults
and whispercpp_advanced_persistence tests
Fixes#363
* fix(whispercpp): gracefully filter unsupported params for pywhispercpp 1.4.1
pywhispercpp 1.4.1 does not expose no_timestamps or suppress_non_speech_tokens
to Python even though they exist in the underlying whisper.cpp C struct.
This causes AttributeError on startup.
Add _load_model_with_compatible_params() which catches AttributeError,
identifies the unsupported param from the error message, removes it, and
retries until the model loads successfully or a non-AttributeError occurs.
Also refactor _build_whispercpp_model_kwargs() from _load_whispercpp_model()
for testability.
Fixes local startup crash reported during PR testing.
* fix(whispercpp): persist advanced settings correctly
* chore(config): add advanced power-user default
Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent)
* feat(settings): gate advanced controls behind power-user opt-in
Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent)
* fix(whispercpp): prefilter unsupported native params
* fix(settings): polish advanced whispercpp controls
* fix(installer): reuse existing whispercpp builds (#421)
* docs: clarify PyPI installation requirements (#423)
* chore(deps): bump next (#429)
Bumps the npm_and_yarn group with 1 update in the /web directory: [next](https://github.com/vercel/next.js).
Updates `next` from 15.5.15 to 15.5.18
- [Release notes](https://github.com/vercel/next.js/releases)
- [Changelog](https://github.com/vercel/next.js/blob/canary/release.js)
- [Commits](https://github.com/vercel/next.js/compare/v15.5.15...v15.5.18)
---
updated-dependencies:
- dependency-name: next
dependency-version: 15.5.18
dependency-type: direct:production
dependency-group: npm_and_yarn
...
Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
* fix(installer): repair pywhispercpp library loading (#433)
* fix: play stop sound immediately on release (#426)
* fix(recognition): play stop sound after audio thread joins to prevent capture on short recordings (#436)
Moves play_stop_sound() to after audio_thread.join() but before the
final buffer processing. This closes the race where the audio thread's
in-flight chunk could capture the stop cue for short (≤ ~192 ms) recordings,
while still feeling immediate compared to the original placement after all
recognition cleanup.
* fix(install): correct openSUSE Tumbleweed dependencies (#420)
* fix(install): correct openSUSE Tumbleweed dependencies
* style: format openSUSE installer changes
---------
Co-authored-by: Jatin K Malik <jatinkrmalik@gmail.com>
* fix(installer): harden Debian compatibility layer (#437)
Add missing pywhispercpp build prerequisites (libssl-dev, autoconf,
automake, libtool, patchelf) to both Debian and Ubuntu apt package lists.
These are not pulled in transitively on clean Debian installs and cause
CMake bootstrap failures during source compilation.
Add runtime smoke test for pywhispercpp to verify_installation() so
"installs but does not run" failures surface at install time with
actionable diagnostic steps instead of silently at first launch.
Treat Debian as a first-class target — remove the "primarily designed for
Ubuntu" warning for Debian systems. Remain conservative for non-Debian,
non-Ubuntu distros.
Add DEBIAN_FRONTEND=noninteractive to all apt install commands to prevent
interactive debconf prompts from blocking unattended installs.
Improve ydotool failure messaging on Debian with compile-from-source
instructions since ydotool is not packaged in Debian standard repos.
Add Debian-specific troubleshooting section to DISTRO_COMPATIBILITY.md
covering build deps, ydotool compilation, scoped source builds, and
libwhisper.so resolution diagnostics.
References: https://medium.com/@cslev/talking-to-my-linux-box-without-talking-to-the-cloud-vocalinux-on-debian-without-the-tears-10bf053ea21b
* fix(install): refresh ldconfig after openSUSE typelib install, clarify python3XY placeholder convention (#438)
- Add 'sudo ldconfig' after zypper install in suse_install_appindicator_runtime()
to ensure GI typelib cache is refreshed so newly installed typelib packages
are immediately discoverable by GObject introspection
- Add section-level comment in distro-package-map.yaml explaining that
python3XY is a placeholder convention resolved at install time by
suse_python_package_prefix() / suse_python_package_candidates()
Follow-up polish from review of #420.
* fix(whispercpp): reduce CPU threads and ensure GPU backend builds in dev mode (#439)
* fix(whispercpp): reduce CPU threads and ensure GPU backend builds in dev mode
- Reduce default n_threads from cpu_count() (16) to max 8 (CPU) / max 4 (GPU)
- Add runtime pywhispercpp GPU capability detection (_detect_pywhispercpp_gpu_backend)
- Make n_threads configurable via whispercpp_n_threads config option
- Fix install.sh regression: GPU source build was skipped in --dev mode
- Extract GPU install logic into install_whispercpp_with_gpu_support()
- Add is_pywhispercpp_gpu_capable() to detect GPU libs in installed pywhispercpp
- Force rebuild pywhispercpp when GPU is detected but lacks GPU support
Fixes CPU spiking to 100% on GPU systems due to excessive thread count
and ensures dev installs get GPU acceleration.
* fix(audio): skip output-only HDMI devices when selecting input device
Add _resolve_valid_input_device() that filters devices by
maxInputChannels > 0, falling back preferred → default → first valid.
Previously, HDMI output devices (0 input channels) could be selected
as the default audio input, causing 'Invalid number of channels' errors.
Now both _record_audio() and _attempt_audio_reconnection() validate the
device has input channels before probing or opening the stream.
* fix(whispercpp): treat n_threads=0 as auto-detect, not literal zero
Config default whispercpp_n_threads=0 was being passed directly
to whisper.cpp as n_threads=0, causing 'cannot create std::vector
larger than max_size()' transcription errors. Now 0 and negative
values are treated as auto-detect, falling through to the smart
thread count logic (max 4 with GPU, max 8 on CPU).
* test: add whispercpp_n_threads to mock assertions and config defaults
Update test_main.py mock to include the new whispercpp_n_threads=0 kwarg
and test_config_manager.py to assert the default config value.
* fix(audio): harden device enumeration against test mocks
The new HDMI-filter helper and GPU-backend detector crashed unit tests
that pass MagicMock() for the PyAudio/pathlib surfaces:
- _find_pywhispercpp_shared_library_dirs hit TypeError when tests
monkey-patch os.stat (Path.is_dir internally calls S_ISDIR on a
non-int st_mode). Widen the catch to OSError/TypeError/ValueError so
unconfigured stat results just skip the candidate dir.
- _resolve_valid_input_device crashed with 'MagicMock > int' on
audio.get_device_count() / info.get('maxInputChannels'). Coerce the
count, treat zero/unknown as 'use system default', and include
non-dict info objects rather than filtering everything out.
No production behavior changes: real PyAudio always returns int counts
and dict device info, so the HDMI filter still works as designed.
* test(recognition): cover new HDMI filter and GPU backend detector
- TestResolveValidInputDevice: 9 cases for the HDMI-filter resolver
(preferred passthrough, HDMI fallback, no-default fallback, no-input
None return, and three mock-tolerance escape hatches).
- TestDetectPywhispercppGpuBackend: 4 cases for runtime GPU library
detection (cpu/vulkan/cuda/empty dirs).
- TestFindPywhispercppSharedLibraryDirsResilience: proves the widened
exception catch survives a TypeError from pathlib.Path.is_dir.
* test(text-injection): wait for IBus init thread to avoid race on slow runners
The IBus environment promotion runs in a background daemon thread spawned
by TextInjector.__init__. The two TestIBusRuntimeFallback cases asserted
on injector.environment immediately after construction, which raced the
thread on Python 3.10 GHA runners and produced intermittent failures.
Join the init thread (with a generous timeout) before asserting, so the
test exercises the post-warmup state deterministically.
* chore(release): prepare v0.11.0-beta (#440)
- Bump version from 0.10.2-beta to 0.11.0-beta
- Update all version references across codebase
- Add What's New section for v0.11.0-beta in README and docs/UPDATE.md
- Add v0.11.0-beta entry to website changelog
- Update SECURITY.md supported versions to 0.11.x
- Update website schema, badge, and package version
- Update RELEASE_PROCESS.md version history table
* fix(startup): allow launch without pynput backend (#448)
* feat(vad): add Silero VAD and drop silence-only buffers (#447)
* feat(vad): replace amplitude-based VAD with Silero neural network
Replace the simple amplitude threshold VAD with Silero VAD, a lightweight
neural network model that classifies audio as speech/non-speech. This
addresses the root cause of #302 (silence not stopping recording) and
- Add silero_vad.py wrapper using ONNX Runtime with bundled Silero v5 model
- Accumulate 512-sample chunks with 64-sample context window per official API
- Map existing vad_sensitivity (1-5) to Silero thresholds (0.8-0.3)
- Use max() across sub-chunks to avoid missing speech onsets
- Fall back to amplitude VAD when onnxruntime is unavailable
- Add onnxruntime as optional dependency (pip install vocalinux[vad])
Closes#344
* address review: add tests, LICENSE, logging, validation fixes
- Add unit tests for SileroVAD (process, reset, fallback, sensitivity mapping)
- Add MIT LICENSE for bundled Silero VAD model in data/
- Log which VAD backend is active on init
- Validate chunk length in SileroVAD.process()
- Skip is_speech check until a full Silero chunk is processed
- Add thread safety note to SileroVAD docstring
- Document resampling rate effect on VAD resolution
- Comment explaining why load_silero_vad() catches broad Exception
* address review: bundle license, document [vad] extra, surface backend in UI
- Pack data/LICENSE into the wheel (pyproject package-data) so attribution
travels with the model
- README: short note on enabling neural VAD via vocalinux[vad]
- Settings: VAD Sensitivity row subtitle shows the active backend; if
onnxruntime is missing, an info-row beneath the group spells out the
install command and the restart requirement
- silero_vad.is_silero_available() probes onnxruntime + model presence
w/o instantiating an InferenceSession
- CI installs .[dev,vad] so Silero paths get exercised
- Integration test for _record_audio Silero branch (mocked pyaudio +
scripted SileroVAD), covering silence flush, speech reset, and
sensitivity threshold mapping at sens=1 vs sens=5
- Tests no longer skip when other modules globally mock numpy: pop the
top-level numpy ref so a fresh import gives the real package, and
swap sys.modules["numpy"] inside the _record_audio call so the lazy
import inside the function picks up real numpy too
* fix(vad): drop silence-only recording buffers
* test(vad): align stop buffer expectations
* test(vad): preserve real numpy module
* fix(install): install vad support opportunistically
* test(vad): cover recognition manager vad edges
---------
Co-authored-by: Sbenazar <127324689+Sbenazar@users.noreply.github.com>
* fix(web): clarify speech demo browser support (#449)
* fix black format check
---------
Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: jay <sevenjay@users.noreply.github.com>
Co-authored-by: Jatin K Malik <jatinkrmalik@gmail.com>
Co-authored-by: Fakhareddine <fe.sabbari@claranet.com>
Co-authored-by: Sbenazar <127324689+Sbenazar@users.noreply.github.com>
Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
Co-authored-by: parko636 <alex@alexparkinson.net>
Co-authored-by: juanfradb <94815906+juanfradb@users.noreply.github.com>
Co-authored-by: Pep Locanto <peplocanto@gmail.com>
* feat(vad): replace amplitude-based VAD with Silero neural network
Replace the simple amplitude threshold VAD with Silero VAD, a lightweight
neural network model that classifies audio as speech/non-speech. This
addresses the root cause of #302 (silence not stopping recording) and
- Add silero_vad.py wrapper using ONNX Runtime with bundled Silero v5 model
- Accumulate 512-sample chunks with 64-sample context window per official API
- Map existing vad_sensitivity (1-5) to Silero thresholds (0.8-0.3)
- Use max() across sub-chunks to avoid missing speech onsets
- Fall back to amplitude VAD when onnxruntime is unavailable
- Add onnxruntime as optional dependency (pip install vocalinux[vad])
Closes#344
* address review: add tests, LICENSE, logging, validation fixes
- Add unit tests for SileroVAD (process, reset, fallback, sensitivity mapping)
- Add MIT LICENSE for bundled Silero VAD model in data/
- Log which VAD backend is active on init
- Validate chunk length in SileroVAD.process()
- Skip is_speech check until a full Silero chunk is processed
- Add thread safety note to SileroVAD docstring
- Document resampling rate effect on VAD resolution
- Comment explaining why load_silero_vad() catches broad Exception
* address review: bundle license, document [vad] extra, surface backend in UI
- Pack data/LICENSE into the wheel (pyproject package-data) so attribution
travels with the model
- README: short note on enabling neural VAD via vocalinux[vad]
- Settings: VAD Sensitivity row subtitle shows the active backend; if
onnxruntime is missing, an info-row beneath the group spells out the
install command and the restart requirement
- silero_vad.is_silero_available() probes onnxruntime + model presence
w/o instantiating an InferenceSession
- CI installs .[dev,vad] so Silero paths get exercised
- Integration test for _record_audio Silero branch (mocked pyaudio +
scripted SileroVAD), covering silence flush, speech reset, and
sensitivity threshold mapping at sens=1 vs sens=5
- Tests no longer skip when other modules globally mock numpy: pop the
top-level numpy ref so a fresh import gives the real package, and
swap sys.modules["numpy"] inside the _record_audio call so the lazy
import inside the function picks up real numpy too
* fix(vad): drop silence-only recording buffers
* test(vad): align stop buffer expectations
* test(vad): preserve real numpy module
* fix(install): install vad support opportunistically
* test(vad): cover recognition manager vad edges
---------
Co-authored-by: Sbenazar <127324689+Sbenazar@users.noreply.github.com>
- Bump version from 0.10.2-beta to 0.11.0-beta
- Update all version references across codebase
- Add What's New section for v0.11.0-beta in README and docs/UPDATE.md
- Add v0.11.0-beta entry to website changelog
- Update SECURITY.md supported versions to 0.11.x
- Update website schema, badge, and package version
- Update RELEASE_PROCESS.md version history table
* fix(whispercpp): reduce CPU threads and ensure GPU backend builds in dev mode
- Reduce default n_threads from cpu_count() (16) to max 8 (CPU) / max 4 (GPU)
- Add runtime pywhispercpp GPU capability detection (_detect_pywhispercpp_gpu_backend)
- Make n_threads configurable via whispercpp_n_threads config option
- Fix install.sh regression: GPU source build was skipped in --dev mode
- Extract GPU install logic into install_whispercpp_with_gpu_support()
- Add is_pywhispercpp_gpu_capable() to detect GPU libs in installed pywhispercpp
- Force rebuild pywhispercpp when GPU is detected but lacks GPU support
Fixes CPU spiking to 100% on GPU systems due to excessive thread count
and ensures dev installs get GPU acceleration.
* fix(audio): skip output-only HDMI devices when selecting input device
Add _resolve_valid_input_device() that filters devices by
maxInputChannels > 0, falling back preferred → default → first valid.
Previously, HDMI output devices (0 input channels) could be selected
as the default audio input, causing 'Invalid number of channels' errors.
Now both _record_audio() and _attempt_audio_reconnection() validate the
device has input channels before probing or opening the stream.
* fix(whispercpp): treat n_threads=0 as auto-detect, not literal zero
Config default whispercpp_n_threads=0 was being passed directly
to whisper.cpp as n_threads=0, causing 'cannot create std::vector
larger than max_size()' transcription errors. Now 0 and negative
values are treated as auto-detect, falling through to the smart
thread count logic (max 4 with GPU, max 8 on CPU).
* test: add whispercpp_n_threads to mock assertions and config defaults
Update test_main.py mock to include the new whispercpp_n_threads=0 kwarg
and test_config_manager.py to assert the default config value.
* fix(audio): harden device enumeration against test mocks
The new HDMI-filter helper and GPU-backend detector crashed unit tests
that pass MagicMock() for the PyAudio/pathlib surfaces:
- _find_pywhispercpp_shared_library_dirs hit TypeError when tests
monkey-patch os.stat (Path.is_dir internally calls S_ISDIR on a
non-int st_mode). Widen the catch to OSError/TypeError/ValueError so
unconfigured stat results just skip the candidate dir.
- _resolve_valid_input_device crashed with 'MagicMock > int' on
audio.get_device_count() / info.get('maxInputChannels'). Coerce the
count, treat zero/unknown as 'use system default', and include
non-dict info objects rather than filtering everything out.
No production behavior changes: real PyAudio always returns int counts
and dict device info, so the HDMI filter still works as designed.
* test(recognition): cover new HDMI filter and GPU backend detector
- TestResolveValidInputDevice: 9 cases for the HDMI-filter resolver
(preferred passthrough, HDMI fallback, no-default fallback, no-input
None return, and three mock-tolerance escape hatches).
- TestDetectPywhispercppGpuBackend: 4 cases for runtime GPU library
detection (cpu/vulkan/cuda/empty dirs).
- TestFindPywhispercppSharedLibraryDirsResilience: proves the widened
exception catch survives a TypeError from pathlib.Path.is_dir.
* test(text-injection): wait for IBus init thread to avoid race on slow runners
The IBus environment promotion runs in a background daemon thread spawned
by TextInjector.__init__. The two TestIBusRuntimeFallback cases asserted
on injector.environment immediately after construction, which raced the
thread on Python 3.10 GHA runners and produced intermittent failures.
Join the init thread (with a generous timeout) before asserting, so the
test exercises the post-warmup state deterministically.
- Add 'sudo ldconfig' after zypper install in suse_install_appindicator_runtime()
to ensure GI typelib cache is refreshed so newly installed typelib packages
are immediately discoverable by GObject introspection
- Add section-level comment in distro-package-map.yaml explaining that
python3XY is a placeholder convention resolved at install time by
suse_python_package_prefix() / suse_python_package_candidates()
Follow-up polish from review of #420.
Add missing pywhispercpp build prerequisites (libssl-dev, autoconf,
automake, libtool, patchelf) to both Debian and Ubuntu apt package lists.
These are not pulled in transitively on clean Debian installs and cause
CMake bootstrap failures during source compilation.
Add runtime smoke test for pywhispercpp to verify_installation() so
"installs but does not run" failures surface at install time with
actionable diagnostic steps instead of silently at first launch.
Treat Debian as a first-class target — remove the "primarily designed for
Ubuntu" warning for Debian systems. Remain conservative for non-Debian,
non-Ubuntu distros.
Add DEBIAN_FRONTEND=noninteractive to all apt install commands to prevent
interactive debconf prompts from blocking unattended installs.
Improve ydotool failure messaging on Debian with compile-from-source
instructions since ydotool is not packaged in Debian standard repos.
Add Debian-specific troubleshooting section to DISTRO_COMPATIBILITY.md
covering build deps, ydotool compilation, scoped source builds, and
libwhisper.so resolution diagnostics.
References: https://medium.com/@cslev/talking-to-my-linux-box-without-talking-to-the-cloud-vocalinux-on-debian-without-the-tears-10bf053ea21b
Moves play_stop_sound() to after audio_thread.join() but before the
final buffer processing. This closes the race where the audio thread's
in-flight chunk could capture the stop cue for short (≤ ~192 ms) recordings,
while still feeling immediate compared to the original placement after all
recognition cleanup.
* feat(settings): add Advanced tab with whisper.cpp anti-hallucination parameters
Add a new Advanced settings tab that exposes whisper.cpp decoding
parameters to help users reduce hallucinations and tune recognition
behavior.
Parameters exposed:
- No Timestamps (default: on)
- Suppress Non-Speech Tokens (default: on)
- No Context (default: on)
- Initial Prompt (default: empty)
- Temperature (default: 0.0)
- Temperature Increment (default: -1.0, disables fallback)
- Entropy Threshold (default: 2.4)
- Logprob Threshold (default: -1.0)
- No-Speech Threshold (default: 0.6)
All parameters are saved under the advanced config section and are
only active when the whisper.cpp engine is selected. The UI grays out
these settings and shows an explanatory note when another engine is
active.
Changes:
- config_manager.py: add whisper.cpp keys to DEFAULT_CONFIG[advanced]
- recognition_manager.py: read params from kwargs/config, pass to Model()
- main.py: pass advanced settings from config to SpeechRecognitionManager
- settings_dialog.py: new Advanced tab with switches, spin buttons, entry
- tests: update test_main.py assertions, add whispercpp_advanced_defaults
and whispercpp_advanced_persistence tests
Fixes#363
* fix(whispercpp): gracefully filter unsupported params for pywhispercpp 1.4.1
pywhispercpp 1.4.1 does not expose no_timestamps or suppress_non_speech_tokens
to Python even though they exist in the underlying whisper.cpp C struct.
This causes AttributeError on startup.
Add _load_model_with_compatible_params() which catches AttributeError,
identifies the unsupported param from the error message, removes it, and
retries until the model loads successfully or a non-AttributeError occurs.
Also refactor _build_whispercpp_model_kwargs() from _load_whispercpp_model()
for testability.
Fixes local startup crash reported during PR testing.
* fix(whispercpp): persist advanced settings correctly
* chore(config): add advanced power-user default
Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent)
* feat(settings): gate advanced controls behind power-user opt-in
Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent)
* fix(whispercpp): prefilter unsupported native params
* fix(settings): polish advanced whispercpp controls
The suse case in install_system_dependencies() previously installed all
packages blindly, causing the entire install to fail if a single package
was unavailable (e.g., glslang on some openSUSE repos).
Changes:
- Check each package individually with rpm -q before installing
- Add graceful fallback for glslang -> glslang-devel
- Add graceful fallback for libappindicator-gtk3 -> typelib-1_0-AppIndicator3-0_1 -> libayatana-appindicator3-1
- Continue installation with warning instead of hard failure
Fixes#407
* test(recognition): cover _perform_recognition and _enqueue_audio_segment
The previous commit demoted logger.info('DEBUG: ...') to logger.debug(...)
and removed a dead loop body in _perform_recognition. Codecov flagged 12
uncovered lines — all inside _perform_recognition and _enqueue_audio_segment,
which existing tests always mocked out.
Add direct (non-mocked) tests for both methods:
- _enqueue_audio_segment:
* empty buffer → early return
* normal buffer → enqueues correctly
* full queue → drops oldest and retries
- _perform_recognition:
* processes a segment then exits on None signal
* drains remaining items on None signal
* exits cleanly when idle with empty queue
* continues looping on queue timeout while recording
These tests exercise the actual threading/queue logic and cover the
logger.debug / logger.warning calls as a side effect.
* test(conftest): suppress real desktop notifications during test runs
_show_notification shells out to notify-send, which pops up real system
notifications when tests exercise the model-not-ready path. Add an autouse
fixture that patches _show_notification for all tests EXCEPT those whose
name contains 'show_notification' (so direct tests of the helper still
verify subprocess.Popen calls).
* chore(gitignore): ignore .sisyphus directory
* fix(ibus): wait for engine readiness before first injection
Add a startup readiness probe that pings the IBus engine after activation so first-use dictation no longer races do_enable(). Also switch NO_ENGINE retry behavior to exponential backoff and cover the new flow in tests.
Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent)
Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>
* test(ibus): cover readiness probe edge cases
Add focused tests for the new startup readiness path, including direct PING success, retry exhaustion, unexpected responses, and engine re-activation before injection. This raises coverage on vocalinux.text_injection.ibus_engine from 68% to 73% in the targeted suite.
Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent)
Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>
* fix(text-injection): avoid blocking IBus and clipboard fallbacks
Warm IBus in the background so startup no longer stalls while readiness probing, and make clipboard helpers fail fast with per-session health caching so non-ASCII ydotool fallback does not lock dictation.
Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent)
Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>
* test(ibus): update retry tests for merged NO_ENGINE recovery logic
---------
Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>
* fix(ibus): retry and recover from transient socket failures
Harden IBus injection against connection-refused, timeout, and engine process races so dictation is more resilient in real-world desktop sessions.
Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent)
* fix(text-injection): fallback to non-IBus backend on runtime IBus failures
Allow dictation to continue when IBus runtime injection fails by automatically switching to xdotool/wtype/ydotool fallbacks instead of returning an immediate failure.
Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent)
* fix(ibus): preserve live socket injection during recovery
Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent)
* test(coverage): add runtime fallback and ibus retry coverage
Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent)
* test(ibus): fix do_destroy super() test failing in CI without IBus installed
Patch builtins.super instead of relying on mock_ibus.Engine.do_destroy,
which fails with issubclass() TypeError when VocalinuxEngine is built
with object as its base (IBus unavailable in CI).
* test(ibus): fix MagicMock.__new__ TypeError on Python < 3.13
VocalinuxEngine inherits from MagicMock at import time (IBus is mocked
via sys.modules at test-file level). Calling VocalinuxEngine.__new__(VocalinuxEngine)
delegates to MagicMock.__new__ which runs issubclass(cls, AsyncMockMixin).
On Python < 3.13 that raises TypeError when the module-cached class is a
MagicMock spec object rather than a real type.
Fix: introduce _make_engine() helper that allocates via a thin concrete
subclass whose __new__ calls object.__new__ directly, bypassing the
MagicMock allocation path entirely. Verified across all three
TestVocalinuxEngineDestroy test methods.
* test(ibus): use unbound do_destroy call to avoid MagicMock interference
The previous fix (thin _ConcreteEngine subclass) still failed because
VocalinuxEngine inherits MagicMock.__getattribute__, which intercepts
.do_destroy on instances and returns a mock attribute rather than the
real method - regardless of Python version.
Call VocalinuxEngine.do_destroy(engine) directly with a plain object()
as the self argument instead. do_destroy only uses self for identity
comparison (is VocalinuxEngine._active_instance), so any object works.
This bypasses MagicMock's MRO entirely.
* test(ibus): reload module and instantiate real engine class for do_destroy tests
* refactor(ibus): extract destroy handler for stable cross-version tests
* test(ibus): cover destroy helper branches and do_destroy delegation
* test(ibus): stabilize do_destroy CI behavior across environments
Avoid brittle super() mocking in no-IBus environments and gate the IBus-only parent destroy path test behind runtime capability checks to prevent false CI failures on Python matrix runners.
* test(ibus): remove flaky do_destroy class-path assertions
Drop brittle class-method destroy tests that fail under mocked GI/IBus on older Python runners and keep deterministic helper-level coverage for destroy behavior.
.opentree/ contains git worktrees created by the opentree CLI tool
(https://github.com/axelgar/opentree). These are local developer
artifacts and should not be tracked.
opentree.toml is the per-repo opentree config - also local/personal,
not something contributors need to share.
Override do_destroy() to clear _active_instance when IBus destroys the
engine on keyboard layout switch (Super+Space). Add retry logic in
inject_text() — on NO_ENGINE, re-activates via switch_engine() and
retries once to recover cleanly.
Co-authored-by: Sbenazar <127324689+Sbenazar@users.noreply.github.com>
Replace hard-coded stop tail trim with configurable stop_sound_guard_ms
(default 200ms) so quick key release no longer clips the final spoken
word in push-to-talk mode.
Co-authored-by: Fakhareddine <fe.sabbari@claranet.com>
Fix false local repo detection when users run `uv init` before installation. Changed detection from naive file existence check to content validation:
- For pyproject.toml: check for `name = "vocalinux"`
- For setup.py: check for "vocalinux" string
- Falls through to remote clone mode if neither matches
Fixes#396
- Add Python 3.14 to classifiers
- Add lxml>=6.1.0 to dependencies to fix install on Python 3.14 (CachyOS, Arch)
- Update black target-version to include py314
Fixes#404Closes#405
- Bump version from 0.10.1-beta to 0.10.2-beta in version.py
- Update README.md What's New section with v0.10.2 highlights and bug fixes
- Update SECURITY.md supported versions table
- Update docs/UPDATE.md with v0.10.2 release notes
- Update docs/RELEASE_PROCESS.md version history table
- Update web/src/app/page.tsx softwareVersion and version badge
- Update web/src/app/changelog/page.tsx with v0.10.2 release entry
- Update web/package.json and web/package-lock.json to 0.10.2
Key changes in this release:
- Non-ASCII text injection via ydotool clipboard fallback (#376)
- IBus detection on Wayland without legacy env vars (#381)
- IBus engine process startup fix (#361)
- Missing deps for Pop!_OS and Ubuntu 24.04+ (#379)
- Systematic code quality refactor (#377)
- Redesigned website OG image (#392)
Replace the cartoon-style OG image with a clean, text-based design
inspired by VocaMac's approach:
- Left side: Logo, tagline, feature bullets, GitHub CTA, distro pills
- Right side: Feature cards with SVG icons and colored badges
- Proper SVG vector icons (shield, keyboard, lightning, microphone)
- Brand gradient accent lines top and bottom
- Added SVG source file for future edits
Image size reduced from 1.1MB to 113KB (PNG) and 96KB to 41KB (webp).
- Add Notify mock to global gi.repository mock in conftest.py
- This prevents real desktop notifications during test runs
- Clean up unnecessary comments in test_main_args_deps.py
* fix(text-injection): handle non-ASCII characters with ydotool via clipboard paste (fixes#362)
ydotool simulates evdev key events which only map to US ASCII keycodes.
Non-ASCII characters (accented letters like á, é, ú, ñ, ü, etc.) get
silently dropped during injection, so 'Esdrújula' becomes 'Esdrjula'.
When ydotool is the active Wayland injection tool and the text contains
non-ASCII characters, the new code:
1. Copies text to clipboard via wl-copy (or xclip/xsel fallback)
2. Simulates Ctrl+V via ydotool key to paste it
3. Falls back to regular ydotool type if clipboard tools unavailable
This preserves all Unicode characters perfectly. ASCII-only text still
uses ydotool type directly for lower latency.
* style: fix black formatting in text_injector.py
* test: add coverage for clipboard paste paths (xclip/xsel/wtype fallbacks)
Covers the uncovered paths flagged by Codecov:
- xclip fallback when wl-copy unavailable
- xsel fallback when wl-copy/xclip unavailable
- wtype Ctrl+V paste simulation
- Paste failure (Ctrl+V fails after clipboard copy)
- No clipboard tools available
* refactor(text-injection): eliminate duplication and dead code in clipboard paste path
- _inject_via_clipboard_paste now delegates to _copy_to_clipboard instead
of reimplementing wl-copy/xclip/xsel selection (removes ~30 lines of
duplication identified in code review)
- Remove unreachable wtype branch from _inject_via_clipboard_paste; wtype
handles Unicode natively via Wayland virtual-keyboard protocol so it
never calls this method
- Drop unnecessary time.sleep(0.05) after clipboard write; subprocess.run
for wl-copy is synchronous so the sleep was a no-op
- Add debug log noting the clipboard will be temporarily overwritten so
operators can observe the side-effect without diving into the code
test improvements:
- Fix fragile string-repr assertion in test_ydotool_ascii_uses_type_directly;
now compares actual list slices and verifies wl-copy is NOT called
- Replace test_clipboard_paste_with_wtype with test_wtype_injects_unicode_directly
which actively asserts wtype calls the tool directly without clipboard
- Add test_inject_text_ydotool_non_ascii_end_to_end covering the full
inject_text -> _inject_with_wayland_tool -> _inject_via_clipboard_paste
stack for accented text on Wayland+ydotool (fixes#362)
- Remove now-unnecessary @patch('time.sleep') decorators from three tests
Add a test for the for/else warning path in _setup_engine() introduced
in PR #361: when all 15 socket-readiness poll attempts exhaust without
SOCKET_PATH appearing, a WARNING is logged and activation is attempted
anyway. The new test mocks time.sleep to avoid real delays and asserts
the exact retry count (15) and log message.
Fixes Codecov patch-coverage gap reported on #361.
* fix(ibus): start engine process before checking registration
_setup_engine() gated on is_engine_registered() which queries
`ibus list-engine`. On systems where the IBus daemon doesn't scan
user-level component directories (~/.local/share/ibus/component/),
the engine never appeared in the list, blocking start_engine_process()
from ever being called — even though its register_component() D-Bus
call is the correct registration path for standalone-launched engines.
Reorder _setup_engine() to start the engine process first, relying on
register_component() for runtime registration. Remove the XML
install step from the setup path since install_ibus_component()
triggers `ibus restart` which kills the engine process.
Follow-up to PR #304 which added register_component() support.
* fix(ibus): remove dead code and add engine readiness check
Address review feedback:
- Remove install_ibus_component(), is_engine_registered(),
is_component_up_to_date(), _get_expected_component_xml()
and their tests — no longer called after setup order fix
- Add SOCKET_PATH poll after start_engine_process() to verify
engine is fully initialized before activation
- Rebase onto current main
* feat(web): add v0.10.1 updates and 5 new SEO pages
## New SEO Pages (5)
- /vs-nerd-dictation/ - Vocalinux vs Nerd Dictation comparison
- /whisper-model-guide/ - Whisper model sizes explained
- /voice-typing-vscode/ - How to dictate in VS Code on Linux
- /writers/ - Voice dictation for writers on Linux
- /gnome-kde/ - Voice typing on GNOME vs KDE Plasma
## Existing Page Updates
- homepage: Update JSON-LD featureList with v0.10.1 features, add new FAQs, link new pages in footer
- troubleshooting: Add suspend/resume recovery, tray resources, settings dialog fixes
- shortcuts: Add sound effects toggle, left/right modifier distinction, PTT improvements
- wayland: Add keyboard layout preservation, clipboard fallback note
- compare: Add engine switching safety notes
- faq: Add suspend/resume and keyboard layout FAQs
- alternatives: Link to new Nerd Dictation comparison
- seo-subpage-shell: Add new pages to navigation
- sitemap: Add 5 new URLs
Fixes#257 (SEO expansion)
* fix(web): add theme toggle to subpages, fix dark mode button contrast, simplify footer nav
- Add light/dark theme toggle to SeoSubpageShell header (all subpages)
- Fix duplicate useState/useEffect imports causing build error
- Fix unclosed div in header markup
- Add dark:text-black to CTA buttons across all 15 subpages for contrast on bright green bg
- Add dark:text-black to footer CTA button in SeoSubpageShell
- Simplify subpage footer to mainNavCategories (5 condensed categories) instead of full navCategories
* fix(web): fix list items stacking, add all new pages to nav, restructure footer
- Fix inline-flex → flex on all li elements across 13 pages so bullet points
always stack on separate lines instead of running together horizontally
- Add all 5 new SEO pages to subpage nav mainNavCategories (Whisper Models,
For Writers, VS Code, GNOME vs KDE under Features/Use Cases; new Compare
dropdown with vs Nerd Dictation, Alternatives, Offline, Open Source)
- Add /vs-nerd-dictation/ to homepage footer Learn column
- Restructure homepage footer from 2 crowded columns into 4 lean categorised
columns (Get Started, Learn, Use Cases, Support) — responsive 2-col on
mobile/tablet, 5-col on desktop
Fix two related IBus issues on KDE Plasma Wayland (and similar DEs):
1. IBus not detected without legacy environment variables:
On Wayland, IBus is configured via the DE's Virtual Keyboard setting
and IBus itself recommends unsetting GTK_IM_MODULE / QT_IM_MODULE.
The detection logic only checked these env vars, so it never found
IBus on a properly configured Wayland session.
Fix: When no IM env vars are set, check if ibus-daemon is running
and has an active engine. If another IM (e.g. Fcitx) is explicitly
configured via env vars, respect that and don't override.
2. IBus text injection silently fails on Wayland:
After switching to the Vocalinux IBus engine, the code called
setxkbmap to capture/restore the keyboard layout. On Wayland,
setxkbmap does not reflect the compositor's state and returns
incorrect results (e.g. 'us' when the user has 'de'). Calling
setxkbmap to restore this wrong layout interfered with IBus's
ability to commit text, causing silent injection failures.
Fix: Skip setxkbmap-based XKB layout management entirely on
Wayland sessions. The Wayland compositor manages keyboard layouts
natively, and setxkbmap can only cause harm there.
Fixes#380
Add cmake, libcairo2-dev, and dynamic libgirepository version detection
to fix installation failures on Pop!_OS (Cosmic DE) and newer Ubuntu.
Three issues addressed:
1. cmake: Required by pywhispercpp to build whisper.cpp from source.
Added to all distro package lists (apt, dnf, pacman, zypper, emerge,
apk, xbps, eopkg).
2. libcairo2-dev: Required to build pycairo. Was present in the Debian
package list but missing from the Ubuntu list, which Pop!_OS uses.
3. libgirepository version: Ubuntu 24.04+ and Pop!_OS Cosmic+ ship
libgirepository-2.0-dev instead of libgirepository1.0-dev. Added
runtime detection to select the correct package, mirroring the
existing version-aware handling already in place for Debian.
Reported-by: DavidHoenisch
Ref: https://github.com/jatinkrmalik/vocalinux/discussions/378
* fix(suspend): delay keyboard restart to 6s for USB re-enumeration
The 2s delay is enough for the speech engine, but USB devices
re-enumerate slowly after resume. At T+2s, stale /dev/input nodes
still exist and evdev opens them, but the kernel tears them down
at ~T+4s, causing another round of disconnections.
Split the reinit into two separate timers:
- 2s: speech engine reinit (unchanged)
- 6s: keyboard backend restart (after USB re-enumeration completes)
* fix(suspend): use inotify to detect device settle before keyboard restart
Replace hard-coded 6s keyboard restart delay with event-driven
Gio.File.monitor_directory() on /dev/input. After resume, a 2s settle
timer resets on each inotify event; keyboard shortcuts restart only once
devices stop changing. A 10s safety cap and a 6s fallback timer guard
against environments where inotify is unavailable.
After suspend/resume, evdev devices disconnect and the monitor thread
exits when all file descriptors close. The resume handler only
reinitialized the speech engine, leaving keyboard shortcuts dead.
Restart _setup_keyboard_shortcuts() in _reinit_after_resume() to
re-scan /dev/input and open fresh device file descriptors.
Copy-to-clipboard is now opt-in. Users can enable it in Settings under Text Injection. This avoids unexpected clipboard overwrites on fresh installs and reduces background wl-copy/xclip timeout warnings on Wayland where clipboard access from daemon threads can be unreliable.
Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent)
Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>
* feat(suspend): add D-Bus based suspend/resume handler
Monitors org.freedesktop.login1.Manager.PrepareForSleep via Gio.DBusProxy to detect system suspend/resume. On resume, invokes a callback to reinitialize the audio pipeline and speech model so dictation works immediately without manual intervention.
Fixes#367
Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent)
Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>
* feat(suspend): wire suspend handler into tray indicator and recognition manager
TrayIndicator now creates a SuspendHandler on startup that listens for D-Bus PrepareForSleep signals. On suspend, active recognition is stopped. On resume, a 2-second delay allows audio hardware to stabilize before reinitializing the speech engine and resetting audio reconnection state.
Fixes#367
Ultraworked with [Sisyphus](https://github.com/code-yeongyu/oh-my-openagent)
Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>
---------
Co-authored-by: Sisyphus <clio-agent@sisyphuslabs.ai>
* fix(ptt): prevent premature transcription during push-to-talk silence
In push-to-talk mode, a brief silence while holding the hotkey would
trigger transcription mid-hold, injecting text before key release.
Add recognition_mode parameter to start_recognition() so the silence
timeout in _record_audio() defers enqueuing audio in PTT mode. Audio
accumulates in the buffer until key release calls stop_recognition(),
which already handles final buffer enqueue. Toggle mode is unchanged.
* fix(clipboard): run clipboard copy in background thread to prevent blocking
The _copy_to_clipboard calls (wl-copy, xclip, xsel) have 5s timeouts
each and run synchronously on the recognition thread. When they hang,
they block stop_recognition() from completing, delaying the state
transition to IDLE and preventing the next push-to-talk press.
Move clipboard copy to a daemon thread on the two success paths so it
never blocks transcription or state transitions. The failure-fallback
path stays synchronous since it needs the return value.
