python-voiceio 0.9.2__tar.gz → 0.9.4__tar.gz

{python_voiceio-0.9.2/python_voiceio.egg-info → python_voiceio-0.9.4}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: python-voiceio
-Version: 0.9.2
+Version: 0.9.4
 Summary: Voice dictation for Linux. Speak → text, locally, instantly.
 Author: Hugo Montenegro
 License-Expression: MIT

{python_voiceio-0.9.2 → python_voiceio-0.9.4}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "python-voiceio"
-version = "0.9.2"
+version = "0.9.4"
 description = "Voice dictation for Linux. Speak → text, locally, instantly."
 readme = "README.md"
 license = "MIT"

{python_voiceio-0.9.2 → python_voiceio-0.9.4/python_voiceio.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: python-voiceio
-Version: 0.9.2
+Version: 0.9.4
 Summary: Voice dictation for Linux. Speak → text, locally, instantly.
 Author: Hugo Montenegro
 License-Expression: MIT

{python_voiceio-0.9.2 → python_voiceio-0.9.4}/tests/test_robustness.py

@@ -63,6 +63,7 @@ def _make_recorder():
     cfg.silence_threshold = 0.01
     cfg.silence_duration = 0.6
     cfg.auto_stop_silence_secs = 5.0
+    cfg.auto_stop_no_speech_secs = 20.0
 
     with patch("voiceio.recorder.sd"):
         from voiceio.recorder import AudioRecorder
@@ -72,6 +73,96 @@ def _make_recorder():
     return rec
 
 
+class TestAutoStop:
+    """Auto-stop distinguishes 'spoke then paused' from 'nothing ever heard'."""
+
+    def _feed(self, rec, secs, *, speech, amplitude=0.0):
+        """Drive _callback with `secs` of audio.
+
+        speech       → what the (mocked) VAD reports (is_speech).
+        amplitude    → raw sample level; 0.0 = digital silence (muted mic),
+                       >1e-4 = a real audio signal reaching the mic.
+        """
+        rec._vad.is_speech.return_value = speech
+        frames = int(secs * rec.sample_rate)
+        data = np.full((frames, 1), amplitude, dtype=np.float32)
+        rec._callback(data, frames, None, None)
+
+    def test_muted_mic_fires_no_speech_reason(self):
+        rec = _make_recorder()
+        rec._recording = True
+        fired = []
+        rec.set_on_auto_stop(lambda reason: fired.append(reason))
+        # 21s of digital zeros — a muted / dead mic.
+        self._feed(rec, 21, speech=False, amplitude=0.0)
+        assert fired == ["no_speech"]
+
+    def test_real_audio_vad_misses_does_not_autostop(self):
+        """Regression: real audio the VAD fails to flag as speech must NOT be
+        cut off. This is the 0.9.3 bug — a 20s dictation the Silero VAD scored
+        as silence auto-stopped mid-sentence."""
+        rec = _make_recorder()
+        rec._recording = True
+        fired = []
+        rec.set_on_auto_stop(lambda reason: fired.append(reason))
+        # 25s of clearly-present audio, but the VAD never calls it speech.
+        self._feed(rec, 25, speech=False, amplitude=0.05)
+        assert fired == []  # signal is present → neither auto-stop fires
+
+    def test_no_speech_does_not_fire_early(self):
+        rec = _make_recorder()
+        rec._recording = True
+        fired = []
+        rec.set_on_auto_stop(lambda reason: fired.append(reason))
+        self._feed(rec, 10, speech=False, amplitude=0.0)  # below the 20s threshold
+        assert fired == []
+
+    def test_speech_then_silence_fires_silence_reason(self):
+        rec = _make_recorder()
+        rec._recording = True
+        fired = []
+        rec.set_on_auto_stop(lambda reason: fired.append(reason))
+        self._feed(rec, 1, speech=True)    # heard speech → _heard_speech
+        self._feed(rec, 6, speech=False)   # 6s silence ≥ 5s auto_stop
+        assert fired == ["silence"]
+
+    def test_single_fire(self):
+        """Callback is cleared on fire — a later chunk must not re-trigger."""
+        rec = _make_recorder()
+        rec._recording = True
+        fired = []
+        rec.set_on_auto_stop(lambda reason: fired.append(reason))
+        self._feed(rec, 21, speech=False)
+        self._feed(rec, 21, speech=False)
+        assert fired == ["no_speech"]
+
+
+class TestAutoStopNotification:
+    """The app surfaces a 'no_speech' auto-stop; a plain 'silence' one is quiet."""
+
+    def test_no_speech_muted_mic_notifies(self):
+        vio, _, _ = _make_vio()
+        vio.recorder.has_signal = MagicMock(return_value=False)  # all zeros = muted
+        with patch("voiceio.feedback.notify") as notify:
+            vio._on_auto_stop("no_speech")
+        assert notify.called
+        assert "mute" in " ".join(notify.call_args[0]).lower()
+
+    def test_no_speech_with_signal_notifies_generic(self):
+        vio, _, _ = _make_vio()
+        vio.recorder.has_signal = MagicMock(return_value=True)  # mic works, user quiet
+        with patch("voiceio.feedback.notify") as notify:
+            vio._on_auto_stop("no_speech")
+        assert notify.called
+        assert "mute" not in " ".join(notify.call_args[0]).lower()
+
+    def test_normal_silence_is_quiet(self):
+        vio, _, _ = _make_vio()
+        with patch("voiceio.feedback.notify") as notify:
+            vio._on_auto_stop("silence")
+        assert not notify.called
+
+
 # ===========================================================================
 # 1. stream_health() tests
 # ===========================================================================

python_voiceio-0.9.4/voiceio/__init__.py

@@ -0,0 +1 @@
+__version__ = "0.9.4"

{python_voiceio-0.9.2 → python_voiceio-0.9.4}/voiceio/app.py

@@ -283,8 +283,28 @@ class VoiceIO:
             self._last_hotkey = now
             self._toggle()
 
-    def _on_auto_stop(self) -> None:
-        """Called from audio thread when sustained silence triggers auto-stop."""
+    def _on_auto_stop(self, reason: str = "silence") -> None:
+        """Called from the audio thread when the recorder auto-stops.
+
+        reason="silence": spoke, then went quiet — a normal end of dictation.
+        reason="no_speech": nothing was heard the whole time — a muted mic or a
+        hotkey pressed by accident. Surface it so it isn't a silent failure
+        (today's Discord-mute confusion: recording ran, nothing was captured,
+        and there was no sign why).
+        """
+        if reason == "no_speech":
+            from voiceio import feedback
+            if not self.recorder.has_signal():
+                feedback.notify(
+                    "VoiceIO: microphone is muted",
+                    "Stopped — no audio is reaching voiceio. Check your mic's "
+                    "mute button and that the right input device is selected.",
+                )
+            else:
+                feedback.notify(
+                    "VoiceIO: stopped — no speech detected",
+                    "Recording auto-stopped after silence.",
+                )
         threading.Thread(target=self._request_stop, daemon=True).start()
 
     def _request_stop(self) -> None:

{python_voiceio-0.9.2 → python_voiceio-0.9.4}/voiceio/config.py

@@ -71,6 +71,11 @@ class AudioConfig:
     silence_threshold: float = 0.01
     silence_duration: float = 0.6
     auto_stop_silence_secs: float = 5.0
+    # Safety net for a muted mic or a forgotten hotkey: if NO speech is heard
+    # for this long from the start, stop and notify. The 5s auto-stop above
+    # only fires once speech has been heard, so it can't catch a dead mic.
+    # 0 disables.
+    auto_stop_no_speech_secs: float = 20.0
     vad_backend: str = "silero"  # "silero" | "rms"
     vad_threshold: float = 0.5  # Silero speech probability threshold
 

{python_voiceio-0.9.2 → python_voiceio-0.9.4}/voiceio/recorder.py

@@ -15,6 +15,11 @@ if TYPE_CHECKING:
 
 log = logging.getLogger(__name__)
 
+# Peak amplitude below this is treated as no signal at all (a muted mic emits
+# exact zeros; a working mic's "silence" still noises above ~1e-4). Matches the
+# floor in has_signal(). Used only to detect a dead mic, never speech vs pause.
+_SIGNAL_FLOOR = 1e-4
+
 
 class RingBuffer:
     """Fixed-size ring buffer for float32 audio samples."""
@@ -108,9 +113,16 @@ class AudioRecorder:
 
         # Auto-stop on sustained silence
         self._auto_stop_secs = cfg.auto_stop_silence_secs
+        self._no_speech_secs = cfg.auto_stop_no_speech_secs
         self._sustained_silence = 0.0
         self._heard_speech = False
-        self._on_auto_stop: Callable[[], None] | None = None
+        # Raw-signal (amplitude) tracking for the muted-mic auto-stop, kept
+        # separate from the VAD-based _heard_speech / _sustained_silence above.
+        self._heard_signal = False
+        self._silent_signal_secs = 0.0
+        # Called with a reason: "silence" (spoke then paused) or "no_speech"
+        # (nothing ever heard — a muted mic or a forgotten hotkey).
+        self._on_auto_stop: Callable[[str], None] | None = None
 
         # Heartbeat: updated by _callback, checked by health watchdog
         self._last_callback_time: float = 0.0
@@ -218,6 +230,8 @@ class AudioRecorder:
             self._silent_chunks = 0.0
             self._sustained_silence = 0.0
             self._heard_speech = False
+            self._heard_signal = False
+            self._silent_signal_secs = 0.0
             self._pause_fired_at = 0
             self._meter_peak = 0.0
             self._meter_clipped = 0
@@ -260,7 +274,7 @@ class AudioRecorder:
         """Set/clear the speech pause callback (used by streaming session)."""
         self._on_speech_pause = callback
 
-    def set_on_auto_stop(self, callback: Callable[[], None] | None) -> None:
+    def set_on_auto_stop(self, callback: Callable[[str], None] | None) -> None:
         """Set/clear the auto-stop callback (fires after sustained silence)."""
         self._on_auto_stop = callback
 
@@ -298,8 +312,19 @@ class AudioRecorder:
 
         # Level metering
         abs_chunk = np.abs(chunk.ravel())
-        self._meter_peak = max(self._meter_peak, float(abs_chunk.max(initial=0.0)))
+        chunk_peak = float(abs_chunk.max(initial=0.0))
+        self._meter_peak = max(self._meter_peak, chunk_peak)
         self._meter_samples += len(abs_chunk)
+
+        # Raw-signal presence, independent of the VAD. A muted mic delivers
+        # digital zeros; real audio (even speech the VAD fails to classify)
+        # sits well above the floor. This — NOT the VAD — gates the muted-mic
+        # auto-stop, so a recording is never cut while real audio is coming in.
+        if chunk_peak > _SIGNAL_FLOOR:
+            self._heard_signal = True
+            self._silent_signal_secs = 0.0
+        else:
+            self._silent_signal_secs += frames / self.sample_rate
         pinned = abs_chunk >= 0.99
         if np.count_nonzero(pinned) >= 4:
             runs = np.convolve(pinned.astype(np.int8), np.ones(4, dtype=np.int8), "valid")
@@ -340,4 +365,20 @@ class AudioRecorder:
             self._on_auto_stop = None
             self._sustained_silence = 0.0
             log.info("Auto-stopping after %.0fs of silence", self._auto_stop_secs)
-            cb()
+            cb("silence")
+
+        # Safety net for a dead mic: NO raw audio signal at all for a long
+        # time — the mic is muted / unplugged / delivering zeros, never merely
+        # that the VAD didn't flag speech (that would cut real dictation off
+        # mid-sentence). _heard_signal latches on the first non-zero chunk, so
+        # this only fires for a mic that was silent from the very start.
+        elif (self._on_auto_stop is not None
+                and self._no_speech_secs > 0
+                and not self._heard_signal
+                and self._silent_signal_secs >= self._no_speech_secs):
+            cb = self._on_auto_stop
+            self._on_auto_stop = None
+            self._silent_signal_secs = 0.0
+            log.info("Auto-stopping: no mic signal in %.0fs (muted?)",
+                     self._no_speech_secs)
+            cb("no_speech")

python_voiceio-0.9.2/voiceio/__init__.py

@@ -1 +0,0 @@
-__version__ = "0.9.2"