python-voiceio 0.9.7__tar.gz → 0.9.9__tar.gz

{python_voiceio-0.9.7/python_voiceio.egg-info → python_voiceio-0.9.9}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: python-voiceio
-Version: 0.9.7
+Version: 0.9.9
 Summary: Voice dictation for Linux. Speak → text, locally, instantly.
 Author: Hugo Montenegro
 License-Expression: MIT

{python_voiceio-0.9.7 → python_voiceio-0.9.9}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "python-voiceio"
-version = "0.9.7"
+version = "0.9.9"
 description = "Voice dictation for Linux. Speak → text, locally, instantly."
 readme = "README.md"
 license = "MIT"

{python_voiceio-0.9.7 → python_voiceio-0.9.9/python_voiceio.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: python-voiceio
-Version: 0.9.7
+Version: 0.9.9
 Summary: Voice dictation for Linux. Speak → text, locally, instantly.
 Author: Hugo Montenegro
 License-Expression: MIT

{python_voiceio-0.9.7 → python_voiceio-0.9.9}/tests/test_platform.py

@@ -72,3 +72,34 @@ def test_platform_frozen():
     import pytest
     with pytest.raises(AttributeError):
         p.os = "darwin"
+
+
+class TestOpenInTerminal:
+    """The tray menu launches print-and-exit commands (history/doctor/logs);
+    hold=True must keep the window open instead of flashing it shut."""
+
+    def _capture(self, monkeypatch, *, hold):
+        import voiceio.platform as plat
+        monkeypatch.setattr(plat, "_detect_os", lambda: "linux")
+        monkeypatch.setenv("TERMINAL", "myterm")
+        monkeypatch.setattr(
+            plat.shutil, "which",
+            lambda b: "/usr/bin/myterm" if b == "myterm" else None,
+        )
+        calls = []
+        with patch("subprocess.Popen", side_effect=lambda args, *a, **k: calls.append(args)):
+            ok = plat.open_in_terminal(["voiceio", "history"], hold=hold)
+        assert ok is True
+        return calls[0]
+
+    def test_hold_wraps_in_shell_that_pauses(self, monkeypatch):
+        args = self._capture(monkeypatch, hold=True)
+        assert args[:2] == ["myterm", "-e"]
+        assert "sh" in args and "-c" in args
+        script = args[-1]
+        assert "voiceio history" in script
+        assert "read" in script  # pauses so the window survives
+
+    def test_no_hold_runs_command_directly(self, monkeypatch):
+        args = self._capture(monkeypatch, hold=False)
+        assert args == ["myterm", "-e", "voiceio", "history"]

{python_voiceio-0.9.7 → python_voiceio-0.9.9}/tests/test_postcorrect.py

@@ -121,23 +121,15 @@ def test_disfluency_mode_rejects_rewording():
         assert pc.correct(original) == original
 
 
-def test_disfluency_mode_rejects_dropped_negation():
-    """Dropping 'not' is one deletion under every fraction cap but inverts
-    meaning — the negation guard must reject it."""
+def test_disfluency_mode_allows_trailing_shrug_with_negation():
+    """No mechanical negation guard — the LLM decides. Dropping a trailing
+    'I don't know' shrug (plus other filler) is accepted; the model, not a
+    regex, is trusted to avoid genuine polarity flips (see the prompt)."""
     pc = PostCorrector(_cfg(remove_disfluencies=True))
-    original = "I do not want the feature to ship today at all"
-    inverted = "I do want the feature to ship today at all"
-    with patch("voiceio.llm_api.chat", return_value=inverted):
-        assert pc.correct(original) == original
-
-
-def test_disfluency_mode_rejects_replaced_negation():
-    """Replacing a negation with something else also inverts meaning."""
-    pc = PostCorrector(_cfg(remove_disfluencies=True))
-    original = "we should never deploy on a friday afternoon"
-    changed = "we should always deploy on a friday afternoon"
-    with patch("voiceio.llm_api.chat", return_value=changed):
-        assert pc.correct(original) == original
+    original = "the build passed but I don't know we should still check it carefully today"
+    cleaned = "the build passed we should still check it carefully today"
+    with patch("voiceio.llm_api.chat", return_value=cleaned):
+        assert pc.correct(original) == cleaned
 
 
 def test_fix_mode_still_rejects_big_deletion():

python_voiceio-0.9.9/voiceio/__init__.py

@@ -0,0 +1 @@
+__version__ = "0.9.9"

{python_voiceio-0.9.7 → python_voiceio-0.9.9}/voiceio/platform.py

@@ -311,8 +311,14 @@ def system_deps_install_cmd(keys: list[str] | None = None) -> str:
     return f"{prefix} {' '.join(pkgs)}"
 
 
-def open_in_terminal(cmd: list[str]) -> bool:
-    """Launch a command in the user's terminal emulator. Returns success."""
+def open_in_terminal(cmd: list[str], *, hold: bool = False) -> bool:
+    """Launch a command in the user's terminal emulator. Returns success.
+
+    ``hold=True`` keeps the window open after the command exits — required for
+    print-and-exit commands (history, doctor, logs) that would otherwise flash
+    a terminal that closes the instant they finish.
+    """
+    import shlex
     import subprocess
 
     plat_os = _detect_os()
@@ -334,11 +340,20 @@ def open_in_terminal(cmd: list[str]) -> bool:
             log.warning("Failed to open cmd.exe")
             return False
 
-    # Linux: try common terminal emulators
+    # Linux: try common terminal emulators. When holding, wrap the command in a
+    # shell that pauses at the end so the window survives a print-and-exit.
+    launch = cmd
+    if hold:
+        inner = " ".join(shlex.quote(c) for c in cmd)
+        launch = [
+            "sh", "-c",
+            f'{inner}; printf "\\n[voiceio] Press Enter to close…"; read _',
+        ]
+
     term = os.environ.get("TERMINAL")
     if term and shutil.which(term):
         try:
-            subprocess.Popen([term, "-e", *cmd])
+            subprocess.Popen([term, "-e", *launch])
             return True
         except OSError:
             pass
@@ -346,7 +361,7 @@ def open_in_terminal(cmd: list[str]) -> bool:
     # x-terminal-emulator (Debian/Ubuntu alternative)
     if shutil.which("x-terminal-emulator"):
         try:
-            subprocess.Popen(["x-terminal-emulator", "-e", *cmd])
+            subprocess.Popen(["x-terminal-emulator", "-e", *launch])
             return True
         except OSError:
             pass
@@ -365,7 +380,7 @@ def open_in_terminal(cmd: list[str]) -> bool:
     for prefix, binary in _TERMINALS:
         if shutil.which(binary):
             try:
-                subprocess.Popen([*prefix, *cmd])
+                subprocess.Popen([*prefix, *launch])
                 return True
             except OSError:
                 continue

{python_voiceio-0.9.7 → python_voiceio-0.9.9}/voiceio/postcorrect.py

@@ -34,18 +34,17 @@ _MAX_INSERTED_WORDS = 0     # adding any word = altering meaning → reject
 _MAX_REPLACE_FRAC = 0.15    # ASR word-fixes only, never wholesale rewording
 _MAX_DELETE_FRAC = 0.4      # backstop against deleting real content
 
-# Words whose deletion/replacement flips meaning — the fraction caps can't catch
-# a single dropped "not". If the edit touches any of these on the original side,
-# reject outright. (Contractions ending in "n't" are handled separately.)
-_MEANING_CRITICAL = frozenset({
-    "not", "no", "never", "none", "nor", "neither", "without", "cannot",
-    "nothing", "nobody", "nowhere", "n't",
-})
 
+def _norm_words(text: str) -> list[str]:
+    """Lowercased, punctuation-stripped word list for the guard's diff, so a
+    sentence split or capitalization isn't misread as a content change."""
+    out = []
+    for w in text.split():
+        t = w.lower().strip(".,;:!?\"'()[]{}—–-")
+        if t:
+            out.append(t)
+    return out
 
-def _is_meaning_critical(word: str) -> bool:
-    w = word.lower().strip(".,;:!?\"'()")
-    return w in _MEANING_CRITICAL or w.endswith("n't")
 
 _SYSTEM_PROMPT = (
     "You fix automatic speech recognition errors in dictated text. "
@@ -59,18 +58,27 @@ _SYSTEM_PROMPT = (
 # Disfluency mode: also strip spoken filler, delete-only. The strict rules
 # mirror the guards — the model is told exactly what the diff check enforces.
 _SYSTEM_PROMPT_CLEAN = (
-    "You convert dictated speech into clean written text. The user dictates "
-    "about software engineering and their projects. Do exactly two things:\n"
+    "You convert dictated speech into clean, concise written text. The user "
+    "dictates about software engineering and their projects, and wants it to "
+    "read like something they wrote, not spoke. Do exactly two things:\n"
     "1. Fix words the recognizer misheard (wrong proper nouns, homophones, "
     "garbled technical terms).\n"
-    "2. Remove speech disfluencies: filler sounds (um, uh, er); filler uses of "
-    "'like', 'you know', 'I mean'; false starts and self-corrections (keep the "
-    "corrected version); and stray word repetitions.\n"
-    "STRICT RULES: Only DELETE disfluencies and FIX misheard words. NEVER add "
-    "words. NEVER rephrase, reword, reorder, or summarize. NEVER drop real "
-    "content, meaningful hedges, or negations. If unsure whether something is a "
-    "disfluency, KEEP it. Preserve the speaker's own wording and punctuation. "
-    "Return only the cleaned text, nothing else."
+    "2. Cut spoken disfluency so it reads tight and clear:\n"
+    "   - filler sounds (um, uh, er);\n"
+    "   - filler discourse markers with no content: leading 'so', 'yeah', "
+    "'okay', 'well', 'now', 'right', and 'you know', 'I mean', 'like', 'sort "
+    "of'/'kind of' when used as filler, 'or something'/'or whatever', a "
+    "trailing 'right?' tag;\n"
+    "   - false starts and self-corrections (keep only the corrected version);\n"
+    "   - stray word and phrase repetitions.\n"
+    "Lean toward the briefer version — remove hesitation and padding freely.\n"
+    "HARD LIMITS (these protect meaning): only DELETE and FIX — NEVER add "
+    "words, rephrase, reword, reorder, or summarize. NEVER change what a "
+    "statement asserts — above all never flip its polarity (do NOT turn 'I "
+    "don't think we should ship' into 'we should ship'), and keep hedges that "
+    "carry real uncertainty about a claim ('maybe', 'probably'). A trailing "
+    "verbal shrug like 'I don't know' with nothing after it is filler you may "
+    "drop. Keep the speaker's own wording. Return only the cleaned text."
 )
 
 _MAX_RECENT = 3
@@ -307,11 +315,13 @@ class PostCorrector:
     def _guard(self, text: str, corrected: str) -> tuple[bool, str]:
         """Decide whether the LLM's edit is within bounds.
 
-        Returns (accept, reject_reason). In disfluency mode the promise is
-        "never change meaning", enforced on the word-level diff: zero
-        insertions (nothing added/rephrased), few replacements (ASR fixes, not
-        rewording), bounded deletions (can't strip real content). Otherwise the
-        original conservative fix-only guards apply.
+        Returns (accept, reject_reason). In disfluency mode the LLM is trusted
+        to judge what's filler; the diff only backstops gross failure — nothing
+        ADDED (no rewrite/hallucination) and not more than a bounded fraction
+        deleted or word-swapped. Comparison is on normalized tokens (lowercased,
+        punctuation stripped) so cosmetic restructuring — a sentence split, a
+        capital letter, a comma — is never miscounted as rewording. Outside
+        disfluency mode the original conservative fix-only guards apply.
         """
         aw, bw = text.split(), corrected.split()
         if not self._remove_disfluencies:
@@ -322,18 +332,16 @@ class PostCorrector:
                 return False, "editratio"
             return True, ""
 
+        na, nb = _norm_words(text), _norm_words(corrected)
+        aw, bw = na, nb
         inserted = replaced = deleted = 0
         for tag, i1, i2, j1, j2 in difflib.SequenceMatcher(a=aw, b=bw).get_opcodes():
             if tag == "insert":
                 inserted += j2 - j1
             elif tag == "replace":
                 replaced += max(i2 - i1, j2 - j1)
-                if any(_is_meaning_critical(w) for w in aw[i1:i2]):
-                    return False, "negation"   # e.g. "not" → something else
             elif tag == "delete":
                 deleted += i2 - i1
-                if any(_is_meaning_critical(w) for w in aw[i1:i2]):
-                    return False, "negation"   # dropping "not" inverts meaning
         n = len(aw)
         if inserted > _MAX_INSERTED_WORDS:
             return False, "inserted"          # added content — meaning changed

{python_voiceio-0.9.7 → python_voiceio-0.9.9}/voiceio/tray/__init__.py

@@ -146,7 +146,7 @@ def _read_stdout(proc: subprocess.Popen, toggle_cb: Callable[[], None]) -> None:
                 action = cmd[5:]
                 cli_cmd = _MENU_COMMANDS.get(action)
                 if cli_cmd:
-                    if not open_in_terminal(cli_cmd):
+                    if not open_in_terminal(cli_cmd, hold=True):
                         log.warning("Failed to open terminal for: %s", action)
     except (OSError, ValueError):
         pass

{python_voiceio-0.9.7 → python_voiceio-0.9.9}/voiceio/tray/_pystray.py

@@ -45,7 +45,7 @@ def start(
             from voiceio.platform import open_in_terminal
             cli_cmd = _MENU_COMMANDS.get(action)
             if cli_cmd:
-                open_in_terminal(cli_cmd)
+                open_in_terminal(cli_cmd, hold=True)
         return _handler
 
     _icon = pystray.Icon(

python_voiceio-0.9.7/voiceio/__init__.py

@@ -1 +0,0 @@
-__version__ = "0.9.7"