scribe-cli 1.2.0__tar.gz → 1.2.1__tar.gz

{scribe_cli-1.2.0 → scribe_cli-1.2.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: scribe-cli
-Version: 1.2.0
+Version: 1.2.1
 Summary: Speech-to-text CLI and system-tray app for dictating into any focused window. Local (vosk, faster-whisper) or cloud (groq, openai) backends, batch or streaming.
 Author-email: Mahé Perrette <mahe.perrette@gmail.com>
 License: MIT License

{scribe_cli-1.2.0 → scribe_cli-1.2.1}/docs/cli.md

@@ -65,6 +65,7 @@ flag suppresses only its own side (giving `--prompt ""` still loads
 | `-m, --mode {keystroke,clipboard,terminal,file}` | Where transcribed text goes (default `keystroke`). `file` routes the transcript exclusively to `--output-file` and suppresses keyboard/clipboard output. See [output.md](output.md). |
 | `--typer {auto,eitype,pynput,wtype,ydotool}` | Keystroke-injection backend (default `auto`).                                |
 | `--type-direct`             | In keystroke mode, type the transcription as keystrokes instead of synthesising Ctrl+V.     |
+| `--typer-delay MS`          | With `--type-direct`, milliseconds between key events for eitype/wtype (default `2`). Some apps, such as the Claude Code prompt, drop the end of text typed at 0. |
 | `-o, --output-file FILE`    | Path the transcription is appended to when `--mode file`. Defaults to `<user-desktop>/scribe-notes.txt` (the platform Desktop folder — `~/Desktop` on Linux/macOS, `%USERPROFILE%\Desktop` on Windows; falls back to home dir if Desktop is absent). Ignored when `--mode` is anything other than `file` (the four output modes are mutually exclusive). |
 
 ## Silence detection

{scribe_cli-1.2.0 → scribe_cli-1.2.1}/scribe/_version.py

@@ -18,7 +18,7 @@ version_tuple: tuple[int | str, ...]
 commit_id: str | None
 __commit_id__: str | None
 
-__version__ = version = '1.2.0'
-__version_tuple__ = version_tuple = (1, 2, 0)
+__version__ = version = '1.2.1'
+__version_tuple__ = version_tuple = (1, 2, 1)
 
-__commit_id__ = commit_id = 'gc1c4cb389'
+__commit_id__ = commit_id = 'g7a322c71b'

{scribe_cli-1.2.0 → scribe_cli-1.2.1}/scribe/app.py

@@ -467,6 +467,12 @@ def get_parser():
                             "is the ^V control character. Cost: slower for long text; on wtype/ydotool "
                             "non-ASCII characters fall back to their ASCII equivalents (eitype is "
                             "Unicode-correct).")
+    group.add_argument("--typer-delay", default=2, type=int,
+                       help="With --type-direct, delay in milliseconds between key events "
+                            "for the eitype and wtype typers (default: %(default)s). Some "
+                            "apps drop keystrokes that arrive faster than they can read them "
+                            "(e.g. the Claude Code prompt loses the end of the text at 0). "
+                            "Costs about 2 x delay per character.")
     group.add_argument("-o", "--output-file",
                        default=DEFAULT_OUTPUT_FILE,
                        help=f"Path the transcription is appended to when "
@@ -675,6 +681,7 @@ def _resolve_output(o, *, is_streaming, backend_obj):
         typer=getattr(o, "typer", None) if mode == "keystroke" else None,
         type_direct=getattr(o, "type_direct", False),
         output_file=getattr(o, "output_file", None) if mode == "file" else None,
+        typer_delay=getattr(o, "typer_delay", 0),
         is_streaming=is_streaming,
         backend_obj=backend_obj,
     )

{scribe_cli-1.2.0 → scribe_cli-1.2.1}/scribe/backends/whisper.py

@@ -54,6 +54,7 @@ class WhisperTranscriber(AbstractTranscriber):
         return {"text": text}
 
     def finalize(self):
+        self.flush_trailing_silence()
         if len(self.session.audio_buffer) == 0:
             return {"text": ""}
         result = self.transcribe_audio(self.session.audio_buffer)

{scribe_cli-1.2.0 → scribe_cli-1.2.1}/scribe/backends/whisper_futo.py

@@ -199,6 +199,7 @@ class WhisperFutoTranscriber(AbstractTranscriber):
         return {"text": text}
 
     def finalize(self):
+        self.flush_trailing_silence()
         if len(self.session.audio_buffer) == 0:
             return {"text": ""}
         result = self.transcribe_audio(self.session.audio_buffer)

{scribe_cli-1.2.0 → scribe_cli-1.2.1}/scribe/models.py

@@ -482,6 +482,20 @@ class AbstractTranscriber(STTBackend):
     def clear_streaming_context(self):
         self._streaming_context = ""
 
+    def flush_trailing_silence(self):
+        """Clip mode: move the silence held since the last speech block into
+        audio_buffer before finalize() transcribes it. The silence gate can
+        classify a quiet word ending as silence, and Whisper tends to drop
+        the last word when the audio stops right at the end of speech.
+        silence_buffer is already capped at clip_max_silence seconds, so at
+        most that much is appended. No-op when nothing was spoken, so a
+        silent recording still finalizes to empty text."""
+        session = self.session
+        if self.pseudo_streaming or not session.audio_buffer:
+            return
+        session.audio_buffer += session.silence_buffer
+        session.silence_buffer = b''
+
     def transcribe_audio(self, audio_data):
         raise NotImplementedError()
 

{scribe_cli-1.2.0 → scribe_cli-1.2.1}/scribe/output.py

@@ -194,7 +194,7 @@ class KeyboardOutput(Output):
 
 def make_output(mode: str, *, typer: Optional[str], type_direct: bool,
                 output_file: Optional[str], is_streaming: bool,
-                backend_obj=None) -> Output:
+                backend_obj=None, typer_delay: int = 0) -> Output:
     """Resolve ``(mode, typer, type_direct, output_file, is_streaming)``
     into the right Output subclass.
 
@@ -230,6 +230,8 @@ def make_output(mode: str, *, typer: Optional[str], type_direct: bool,
     if type_direct:
         from scribe.typers import pick_typer
         typer_obj = pick_typer(typer if typer and typer != "auto" else None)
+        if hasattr(typer_obj, "key_delay_ms"):
+            typer_obj.key_delay_ms = typer_delay
     else:
         typer_obj = None
     return KeyboardOutput(

{scribe_cli-1.2.0 → scribe_cli-1.2.1}/scribe/session.py

@@ -168,6 +168,17 @@ class RecordingSession:
                 microphone.q.queue.clear()
                 result = {"text": ""}
             else:
+                # The loop above polls the queue every 100 ms, so blocks
+                # captured just before stop can still be waiting here. Feed
+                # them to batch backends before finalizing so the last words
+                # aren't lost. A pseudo-streaming silence cut at this point
+                # is moot: finalize() runs right after anyway.
+                if not streaming:
+                    try:
+                        while not microphone.q.empty():
+                            self.backend.transcribe_realtime_audio(microphone.q.get())
+                    except SilenceDetected:
+                        pass
                 try:
                     result = self.backend.finalize()
                 except Exception as exc:

{scribe_cli-1.2.0 → scribe_cli-1.2.1}/scribe/typers/eitype.py

@@ -12,6 +12,9 @@ from scribe.typers.base import Typer
 
 class EitypeTyper:
     name = "eitype"
+    # Milliseconds between key events (`eitype -d`). Set from --typer-delay
+    # by scribe.output.make_output; 0 sends the whole string as one burst.
+    key_delay_ms: int = 0
 
     def compatible(self) -> bool:
         """Linux Wayland session; libei is a Wayland-specific protocol."""
@@ -29,7 +32,10 @@ class EitypeTyper:
 
     def _emit(self, text: str) -> None:
         try:
-            subprocess.run(["eitype", "--", text], check=True, capture_output=True)
+            cmd = ["eitype"]
+            if self.key_delay_ms > 0:
+                cmd += ["-d", str(self.key_delay_ms)]
+            subprocess.run(cmd + ["--", text], check=True, capture_output=True)
         except subprocess.CalledProcessError as e:
             raise RuntimeError(
                 f"eitype failed: {e.stderr.decode(errors='replace')}"

{scribe_cli-1.2.0 → scribe_cli-1.2.1}/scribe/typers/wtype.py

@@ -13,6 +13,9 @@ from scribe.typers.base import Typer
 
 class WtypeTyper:
     name = "wtype"
+    # Milliseconds between key events (`wtype -d`). Set from --typer-delay
+    # by scribe.output.make_output; 0 sends the whole string as one burst.
+    key_delay_ms: int = 0
 
     def compatible(self) -> bool:
         """Linux Wayland on a wlroots-based compositor (Sway, Hyprland, …).
@@ -42,7 +45,10 @@ class WtypeTyper:
 
     def _emit(self, text: str) -> None:
         try:
-            subprocess.run(["wtype", "--", text], check=True, capture_output=True)
+            cmd = ["wtype"]
+            if self.key_delay_ms > 0:
+                cmd += ["-d", str(self.key_delay_ms)]
+            subprocess.run(cmd + ["--", text], check=True, capture_output=True)
         except subprocess.CalledProcessError as e:
             raise RuntimeError(
                 f"wtype failed: {e.stderr.decode(errors='replace')}"

{scribe_cli-1.2.0 → scribe_cli-1.2.1}/scribe_cli.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: scribe-cli
-Version: 1.2.0
+Version: 1.2.1
 Summary: Speech-to-text CLI and system-tray app for dictating into any focused window. Local (vosk, faster-whisper) or cloud (groq, openai) backends, batch or streaming.
 Author-email: Mahé Perrette <mahe.perrette@gmail.com>
 License: MIT License

{scribe_cli-1.2.0 → scribe_cli-1.2.1}/scribe_cli.egg-info/SOURCES.txt

@@ -50,6 +50,8 @@ scribe_cli.egg-info/SOURCES.txt
 scribe_cli.egg-info/dependency_links.txt
 scribe_cli.egg-info/entry_points.txt
 scribe_cli.egg-info/requires.txt
+scribe_cli.egg-info/scm_file_list.json
+scribe_cli.egg-info/scm_version.json
 scribe_cli.egg-info/top_level.txt
 scribe_data/__init__.py
 scribe_data/silero_vad.LICENSE

scribe_cli-1.2.1/scribe_cli.egg-info/scm_file_list.json

@@ -0,0 +1,71 @@
+{
+  "files": [
+    ".github/FUNDING.yml",
+    ".github/workflows/docs.yml",
+    ".github/workflows/pypi.yml",
+    ".gitignore",
+    "LICENSE",
+    "README.md",
+    "docs/app-tray-menu.png",
+    "docs/backends.md",
+    "docs/cli.md",
+    "docs/desktop-install.md",
+    "docs/index.md",
+    "docs/installation.md",
+    "docs/output.md",
+    "docs/quickstart.md",
+    "docs/roadmap-libei.md",
+    "docs/tray.md",
+    "icon.xcf",
+    "mkdocs.yml",
+    "pyproject.toml",
+    "scribe/__init__.py",
+    "scribe/app.py",
+    "scribe/audio.py",
+    "scribe/backends/__init__.py",
+    "scribe/backends/groq.py",
+    "scribe/backends/openai_api.py",
+    "scribe/backends/openai_realtime.py",
+    "scribe/backends/vosk.py",
+    "scribe/backends/whisper.py",
+    "scribe/backends/whisper_futo.py",
+    "scribe/dialog.py",
+    "scribe/install_desktop.py",
+    "scribe/keyboard.py",
+    "scribe/menu.py",
+    "scribe/models.py",
+    "scribe/models.toml",
+    "scribe/output.py",
+    "scribe/saverecording.py",
+    "scribe/session.py",
+    "scribe/testpynput.py",
+    "scribe/typers/__init__.py",
+    "scribe/typers/base.py",
+    "scribe/typers/eitype.py",
+    "scribe/typers/pynput.py",
+    "scribe/typers/wtype.py",
+    "scribe/typers/ydotool.py",
+    "scribe/util.py",
+    "scribe_data/__init__.py",
+    "scribe_data/share/icon.png",
+    "scribe_data/share/icon_recording.png",
+    "scribe_data/share/icon_writing.png",
+    "scribe_data/silero_vad.LICENSE",
+    "scribe_data/silero_vad.onnx",
+    "scribe_data/templates/scribe.desktop",
+    "scripts/bench_whisper_local.py",
+    "scripts/test_python_versions_install.sh",
+    "tests/test_backend_matrix.py",
+    "tests/test_clip_silence_trim.py",
+    "tests/test_clipboard_backend.py",
+    "tests/test_compose_prompt.py",
+    "tests/test_debug_logging.py",
+    "tests/test_openai_realtime_coalesce.py",
+    "tests/test_output.py",
+    "tests/test_output_file_picker.py",
+    "tests/test_prompt_file_picker.py",
+    "tests/test_pseudo_streaming.py",
+    "tests/test_typers_ascii_fallback.py",
+    "tests/test_whisper_futo.py"
+  ]
+}

scribe_cli-1.2.1/scribe_cli.egg-info/scm_version.json

@@ -0,0 +1,8 @@
+{
+  "tag": "1.2.1",
+  "distance": 0,
+  "node": "g7a322c71ba43f6e2a90702f6cfaad3e4713c3193",
+  "dirty": false,
+  "branch": "HEAD",
+  "node_date": "2026-10-01"
+}

{scribe_cli-1.2.0 → scribe_cli-1.2.1}/tests/test_clip_silence_trim.py

@@ -90,14 +90,38 @@ def test_short_pause_kept_verbatim():
     assert sess.trimmed_silence_bytes == 0
 
 
-def test_trailing_silence_dropped():
+def test_trailing_silence_capped_at_stop():
     sess = make_session()
     backend = make_backend(sess, clip_max_silence=2.0)
     backend.transcribe_realtime_audio(loud_chunk(1.0))
     for _ in range(50):  # 5 s of trailing silence, never followed by speech
         backend.transcribe_realtime_audio(silent_chunk(0.1))
-    # finalize() reads audio_buffer only — the pause stays out of it.
     assert secs(sess.audio_buffer) == 1.0
+    # At stop, the retained tail (<= clip_max_silence) is appended so a
+    # quiet word ending misread as silence still reaches the backend.
+    backend.flush_trailing_silence()
+    assert secs(sess.audio_buffer) == 3.0
+    assert sess.silence_buffer == b''
+
+
+def test_quiet_word_ending_kept_at_stop():
+    sess = make_session()
+    backend = make_backend(sess, clip_max_silence=2.0)
+    backend.transcribe_realtime_audio(loud_chunk(1.0))
+    # A trailing-off word ending, below the dB gate threshold.
+    quiet = loud_chunk(0.3, amplitude=100)
+    backend.transcribe_realtime_audio(quiet)
+    backend.flush_trailing_silence()
+    assert sess.audio_buffer.endswith(quiet)
+
+
+def test_flush_noop_when_nothing_spoken():
+    sess = make_session()
+    backend = make_backend(sess, clip_max_silence=2.0)
+    for _ in range(10):
+        backend.transcribe_realtime_audio(silent_chunk(0.1))
+    backend.flush_trailing_silence()
+    assert sess.audio_buffer == b''
 
 
 def test_zero_disables_trimming():

{scribe_cli-1.2.0 → scribe_cli-1.2.1}/tests/test_typers_ascii_fallback.py

@@ -78,3 +78,26 @@ def test_unrenderable_char_is_skipped_not_fatal():
 
     type_ascii_safe(emit, "xéy", (LayoutError,))
     assert "".join(typed) == "xy"
+
+
+def _record_subprocess(monkeypatch, module):
+    calls = []
+    monkeypatch.setattr(module.subprocess, "run",
+                        lambda cmd, **kw: calls.append(cmd))
+    return calls
+
+
+def test_subprocess_typers_pass_key_delay(monkeypatch):
+    # --typer-delay reaches eitype/wtype as `-d MS`; some apps (the Claude
+    # Code prompt) drop the end of text typed as one zero-delay burst.
+    import scribe.typers.eitype as eitype_mod
+    import scribe.typers.wtype as wtype_mod
+    for module, cls in ((eitype_mod, eitype_mod.EitypeTyper),
+                        (wtype_mod, wtype_mod.WtypeTyper)):
+        calls = _record_subprocess(monkeypatch, module)
+        typer = cls()
+        typer.type("hi")
+        typer.key_delay_ms = 2
+        typer.type("hi")
+        assert calls == [[cls.name, "--", "hi"],
+                         [cls.name, "-d", "2", "--", "hi"]]