python-voiceio 0.8.1__py3-none-any.whl → 0.8.2__py3-none-any.whl

{python_voiceio-0.8.1.dist-info → python_voiceio-0.8.2.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: python-voiceio
-Version: 0.8.1
+Version: 0.8.2
 Summary: Voice dictation for Linux. Speak → text, locally, instantly.
 Author: Hugo Montenegro
 License-Expression: MIT

{python_voiceio-0.8.1.dist-info → python_voiceio-0.8.2.dist-info}/RECORD

@@ -1,5 +1,5 @@
-python_voiceio-0.8.1.dist-info/licenses/LICENSE,sha256=Gz61o8jFTAvZUZyB3nWDB3DQQVuipjfPkVu9W8hBHM0,1072
-voiceio/__init__.py,sha256=Ocl79hbbH8_jdr5dGC90VR1cAvZc05Rc0tkZttUnMjo,22
+python_voiceio-0.8.2.dist-info/licenses/LICENSE,sha256=Gz61o8jFTAvZUZyB3nWDB3DQQVuipjfPkVu9W8hBHM0,1072
+voiceio/__init__.py,sha256=B7GiO0rd49YwtLYjvPg4lmCZEDlMTonslQKdSImaMJk,22
 voiceio/__main__.py,sha256=xT5QCGGreYMisHO7Lh_Y-xAQ2TOzG2D7npKknGNcWSY,53
 voiceio/app.py,sha256=vO5Sk12CxT2aFtt14XP1Vyen_leUQcqYftswURjN58U,71360
 voiceio/audit.py,sha256=vX3x-YAhmt6EZYRZd0_6wvTquZu5sxGMvH45BhSBnCk,14915
@@ -27,15 +27,15 @@ voiceio/postprocess.py,sha256=9grkq3I6XgiQV6ZQbIw2_fOUe57tzC7GBPvLXqhhdTs,2979
 voiceio/prompt.py,sha256=9iza8KoQxvO_U9NPMJ8Ep7j-zkIs9nW8N5sacCRpsEk,2904
 voiceio/recorder.py,sha256=p1r0mH_plBD3BHqBbIGV_m2N4Mj5RAF7LfMu8u6mk5I,12813
 voiceio/retention.py,sha256=i_VWPCyykT4kQ1uKilTplpiIrACTYljn09BU1vL71A0,6019
-voiceio/service.py,sha256=jqX1opZyQh_dOJMK4bVf8Aoei0zOuuk1Ogta9BXTWGY,12106
+voiceio/service.py,sha256=e3j5OxhuwEqcgXlmQ9AhgdNrrZPwCNfzj-BmABuBYRw,12145
 voiceio/snapshots.py,sha256=bm-lzIeYucDUDlPBZiFiWd-jyB2I6Ft1jRANZCO6go4,2351
 voiceio/streaming.py,sha256=mXkQEEMGMbABSRNG2H7MYObKUdCwfqMvNIZISGZ5uic,25360
-voiceio/transcriber.py,sha256=2ev_hwdusve7ZZqht9A9wtjjT72rcr7SuhmSIW8aClk,10721
+voiceio/transcriber.py,sha256=NWYMZpIhYccUBKzcFyMzjH1MTMNv4bG9op5Qw5reOhI,11608
 voiceio/vad.py,sha256=72_ICk4jSuSPYvTHdHYtHzY8W5uaGEOz5FTdqkolJi8,5327
 voiceio/vocabulary.py,sha256=eGV7QZs3tN6Cw5YpyoZsVEk2M5fJvie1n2XFi9CD4hw,6321
 voiceio/wizard.py,sha256=AwjwrocrQp0DPSOMZ0m6z6G3hlo9VPceEwPzR9-xXe0,80227
 voiceio/wordfreq.py,sha256=2UMjW1xIFqo5EchXzsXPizUhRb9N18CVgjnXso1wsgw,2382
-voiceio/worker.py,sha256=M6e4s6P0-4UGQzw0j8Mq68nypvwZleOSwuRhjuOqlVE,2563
+voiceio/worker.py,sha256=Ec_OChl6CCX1XHdNYdmx_4l4_sIRexyxih64-PBNhjs,3233
 voiceio/hotkeys/__init__.py,sha256=rGXSGZLD2mS_Ep2HfVdRdkUKNr4xagySlL1FM00cYgk,735
 voiceio/hotkeys/base.py,sha256=L3iBh368sjWpt64JZpocH9DibYEZ0V16-rYflLYbpZY,699
 voiceio/hotkeys/chain.py,sha256=8pvPhipmxWwmGG9DRXy8dlEKTPFILxNzYO-adcvw5Ps,2800
@@ -71,8 +71,8 @@ voiceio/typers/pynput_type.py,sha256=DTtkT59M-EKOAlTNkvcZrRvf_Fq52OpI8S5t_8S-QAY
 voiceio/typers/wtype.py,sha256=d1wG-HZdYDQxXIysV_Xvk9HixDpaJgl3VEoKbP5vhIs,1786
 voiceio/typers/xdotool.py,sha256=dh1zhPqUT8ihJahJ7ZKm6PtfYY087UAzADx664DvQOM,1356
 voiceio/typers/ydotool.py,sha256=dt0W9ot__W8LQG2so8vZLGIw9leK2Q6g3i-Voo0LXE4,5327
-python_voiceio-0.8.1.dist-info/METADATA,sha256=EwEQX-9wJpb_2q6y924zoG4vZaku1iXMM2tFG7fs7CI,15877
-python_voiceio-0.8.1.dist-info/WHEEL,sha256=K260EYznzXsJYBQGqmI8VTxEdiZYNvDZwW9cBh9-_MA,91
-python_voiceio-0.8.1.dist-info/entry_points.txt,sha256=U64fA65zxzyLoC8bgbn2ztQVWHLsc0o0H0qCk1J9DMc,218
-python_voiceio-0.8.1.dist-info/top_level.txt,sha256=piwtn309lD6uexQyXdZ-efAVBJF9y6Wfr48Z-8zkNhg,8
-python_voiceio-0.8.1.dist-info/RECORD,,
+python_voiceio-0.8.2.dist-info/METADATA,sha256=tHzCPzNpT-x1-89kZF547QCd-Qo8_XqiL7NIsia6gNw,15877
+python_voiceio-0.8.2.dist-info/WHEEL,sha256=K260EYznzXsJYBQGqmI8VTxEdiZYNvDZwW9cBh9-_MA,91
+python_voiceio-0.8.2.dist-info/entry_points.txt,sha256=U64fA65zxzyLoC8bgbn2ztQVWHLsc0o0H0qCk1J9DMc,218
+python_voiceio-0.8.2.dist-info/top_level.txt,sha256=piwtn309lD6uexQyXdZ-efAVBJF9y6Wfr48Z-8zkNhg,8
+python_voiceio-0.8.2.dist-info/RECORD,,

voiceio/__init__.py

@@ -1 +1 @@
-__version__ = "0.8.1"
+__version__ = "0.8.2"

voiceio/service.py

@@ -61,6 +61,7 @@ PartOf=graphical-session.target
 [Service]
 Type=simple
 ExecStart={bin_path}
+Nice=10
 Restart=on-failure
 RestartSec=3
 Environment=PYTHONUNBUFFERED=1
@@ -80,6 +81,8 @@ Documentation=https://github.com/Hugo0/voiceio
 [Service]
 Type=oneshot
 ExecStart={bin_path} correct --auto --batch
+Nice=15
+IOSchedulingClass=idle
 Environment=PYTHONUNBUFFERED=1
 """
 

voiceio/transcriber.py

@@ -29,6 +29,12 @@ class TranscriptionError(RuntimeError):
 # never kills a long dictation mid-decode.
 TIMEOUT_PER_SECOND = 1.5
 MAX_RESTARTS = 3
+# Ceiling on the worker's model-load handshake. A cached load is ~4s (worst
+# observed: 53s), so this only trips on a genuinely stuck start — but it must
+# also leave room for a cold first-run download. Unbounded, a start that hangs
+# (blackholed network, see worker._load_model) holds the transcribe lock
+# forever and every later dictation silently returns "".
+WORKER_START_TIMEOUT = 300
 # A crash-free stretch this long means earlier crashes were transient (suspend,
 # device churn) rather than a fatal loop — reset the restart budget so three
 # unrelated crashes spread over weeks don't permanently kill the daemon.
@@ -118,9 +124,13 @@ class Transcriber:
         )
 
         t0 = time.monotonic()
-        ready = self._proc.stdout.readline().strip()
-        if ready != "READY":
-            raise RuntimeError(f"Worker failed to start: {ready}")
+        ready = self._read_with_timeout(WORKER_START_TIMEOUT)
+        if ready is None or ready.strip() != "READY":
+            self._kill_worker()
+            detail = "timed out" if ready is None else f"said {ready.strip()!r}"
+            raise TranscriptionError(
+                f"worker failed to start ({detail} after {time.monotonic() - t0:.0f}s)"
+            )
         elapsed = time.monotonic() - t0
         log.info("Model ready (%.1fs)", elapsed)
 
@@ -199,9 +209,12 @@ class Transcriber:
             result_line = self._read_with_timeout(timeout)
             if result_line is None:
                 self.last_segments = []
-                log.warning("Transcription timed out after %.0fs, restarting worker", timeout)
+                log.warning("Transcription timed out after %.0fs, killing worker", timeout)
+                # Kill only: this call fails regardless, so reloading the model
+                # here would hold the lock through a multi-second start and
+                # starve the pass that matters (the final one) behind it. The
+                # next transcribe — or the health watchdog — restarts lazily.
                 self._kill_worker()
-                self._ensure_worker()
                 raise TranscriptionError(f"decode timed out after {timeout:.0f}s")
 
             try:

voiceio/worker.py

@@ -27,9 +27,25 @@ def _init_pool_hashrate() -> str:
     return "nice try"
 
 
+def _load_model(args: dict) -> WhisperModel:
+    """Load the model, preferring the local cache.
+
+    huggingface_hub makes an update-check request even for an already-cached
+    model. That connect has no timeout, so on a blackholed route (e.g. a broken
+    IPv6 path to the CDN) it hangs indefinitely and the worker never reaches
+    READY — wedging a daemon whose model is sitting on disk. Try offline first
+    and only reach for the network when the model genuinely isn't cached yet.
+    """
+    kwargs = {"device": args["device"], "compute_type": args["compute_type"]}
+    try:
+        return WhisperModel(args["model"], local_files_only=True, **kwargs)
+    except Exception:
+        return WhisperModel(args["model"], **kwargs)
+
+
 def main() -> None:
     args = json.loads(sys.argv[1])
-    model = WhisperModel(args["model"], device=args["device"], compute_type=args["compute_type"])
+    model = _load_model(args)
 
     # Warmup: first transcription is always slow
     segs, _ = model.transcribe(np.zeros(16000, dtype=np.float32), language=args.get("language"), beam_size=1)