scribe-cli 0.7.10__tar.gz → 0.7.11__tar.gz

{scribe_cli-0.7.10/scribe_cli.egg-info → scribe_cli-0.7.11}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.2
 Name: scribe-cli
-Version: 0.7.10
+Version: 0.7.11
 Summary: scribe is a local speech recognition tool that provides real-time transcription using vosk and whisper AI, with the goal of serving as a virtual keyboard on a computer
 Author-email: Mahé Perrette <mahe.perrette@gmail.com>
 License: MIT License
@@ -104,7 +104,7 @@ pip install -e .[all]
 You can leave the optional dependencies (leave out `[all]`) but must install at least one of `vosk` or `openai-whisper` packages (see Usage below).
 
 The `vosk` language models will download on-the-fly.
-The default download folder is `$XDG_CACHE_HOME/{backend}` where `$XDG_CACHE_HOME` defaults to `$HOME/.cache` (note for the `whisker` backend
+The default download folder is `$XDG_CACHE_HOME/{backend}` where `$XDG_CACHE_HOME` defaults to `$HOME/.cache` (note for the `whisper` backend
 the default is left to the `openai-whisper` package and might change in the future).
 
 
@@ -122,8 +122,8 @@ You can interrupt the recording via Ctrl + C and start again or change model. Th
 
 The default (`whisper`) is excellent at transcribing a full-length audio sequences in [many languages](https://github.com/openai/whisper?tab=readme-ov-file#available-models-and-languages). It is really impressive,
 but it cannot do real-time, and depending on the model can have relatively long execution time, especially with the `turbo` model (at least on my laptop with CPU only). The `small` model is also excellent and runs much faster. It is selected as default in `scribe` for that reason.
-With the `whisker` model you need to stop the registration manually before the transcription occurs (Ctrl + C), though
-there is a maximum duration after which it will stop by itself, which is setup to 60s by default (unless `--duration` is set to something else).
+With the `whisper` model the registration stops after a 2-second silence is detected. You can also stop the registration manually before the transcription occurs (Ctrl + C or Stop in the `--app` mode).
+By default, the recording will only last 120 seconds. You can fine-tune this behaviour via the `--silence`, `--duration` and `--restart-after-silence` parameters.
 
 The `vosk` backend is much faster and very good at doing real-time transcription for one language, but tended to make more mistakes in my tests and it does not do punctuation.
 It becomes really powerful when used for longer or interactive typing session with the [keyboard](#virtual-keyboard-experimental) option, e.g. to make notes or chat with an AI.

{scribe_cli-0.7.10 → scribe_cli-0.7.11}/README.md

@@ -41,7 +41,7 @@ pip install -e .[all]
 You can leave the optional dependencies (leave out `[all]`) but must install at least one of `vosk` or `openai-whisper` packages (see Usage below).
 
 The `vosk` language models will download on-the-fly.
-The default download folder is `$XDG_CACHE_HOME/{backend}` where `$XDG_CACHE_HOME` defaults to `$HOME/.cache` (note for the `whisker` backend
+The default download folder is `$XDG_CACHE_HOME/{backend}` where `$XDG_CACHE_HOME` defaults to `$HOME/.cache` (note for the `whisper` backend
 the default is left to the `openai-whisper` package and might change in the future).
 
 
@@ -59,8 +59,8 @@ You can interrupt the recording via Ctrl + C and start again or change model. Th
 
 The default (`whisper`) is excellent at transcribing a full-length audio sequences in [many languages](https://github.com/openai/whisper?tab=readme-ov-file#available-models-and-languages). It is really impressive,
 but it cannot do real-time, and depending on the model can have relatively long execution time, especially with the `turbo` model (at least on my laptop with CPU only). The `small` model is also excellent and runs much faster. It is selected as default in `scribe` for that reason.
-With the `whisker` model you need to stop the registration manually before the transcription occurs (Ctrl + C), though
-there is a maximum duration after which it will stop by itself, which is setup to 60s by default (unless `--duration` is set to something else).
+With the `whisper` model the registration stops after a 2-second silence is detected. You can also stop the registration manually before the transcription occurs (Ctrl + C or Stop in the `--app` mode).
+By default, the recording will only last 120 seconds. You can fine-tune this behaviour via the `--silence`, `--duration` and `--restart-after-silence` parameters.
 
 The `vosk` backend is much faster and very good at doing real-time transcription for one language, but tended to make more mistakes in my tests and it does not do punctuation.
 It becomes really powerful when used for longer or interactive typing session with the [keyboard](#virtual-keyboard-experimental) option, e.g. to make notes or chat with an AI.

{scribe_cli-0.7.10 → scribe_cli-0.7.11}/scribe/_version.py

@@ -12,5 +12,5 @@ __version__: str
 __version_tuple__: VERSION_TUPLE
 version_tuple: VERSION_TUPLE
 
-__version__ = version = '0.7.10'
-__version_tuple__ = version_tuple = (0, 7, 10)
+__version__ = version = '0.7.11'
+__version_tuple__ = version_tuple = (0, 7, 11)

{scribe_cli-0.7.10 → scribe_cli-0.7.11}/scribe/app.py

@@ -147,7 +147,8 @@ def get_transcriber(o, prompt=True):
 
     elif backend == "whisper":
         transcriber = WhisperTranscriber(model_name=model, language=o.language, samplerate=o.samplerate,
-                                         timeout=o.duration, silence_duration=o.silence, restart_after_silence=o.restart_after_silence,
+                                         timeout=o.duration, silence_duration=o.silence, silence_thresh=o.silence_db,
+                                         restart_after_silence=o.restart_after_silence,
                                          model_kwargs={"download_root": o.download_folder_whisper})
 
     else:
@@ -178,12 +179,13 @@ def get_parser():
     parser.add_argument("--microphone-device", help="The device index of the microphone to use.", type=int)
     parser.add_argument("--keyboard", action="store_true")
     parser.add_argument("--no-clipboard", dest="clipboard", action="store_false")
-    parser.add_argument("--latency", default=0, type=float, help="keyboard latency")
+    parser.add_argument("--latency", default=0.01, type=float, help="keyboard latency")
     parser.add_argument("--ascii", action="store_true", help="Use unidecode for keyboard typing in ascii")
 
     group = parser.add_argument_group("whisper options")
-    group.add_argument("--duration", default=120, type=int, help="Max duration of the whisper recording (default %(default)ss)")
-    group.add_argument("--silence", default=2, type=float, help="silence duration that prompt transcription (whisper) (default %(default)ss)")
+    group.add_argument("--duration", default=120, type=float, help="Max duration of the whisper recording (default %(default)ss)")
+    group.add_argument("--silence", default=2, type=float, help="silence duration (default %(default)ss)")
+    group.add_argument("--silence-db", default=-30, type=float, help="silence magnitude in db (default %(default)ss)")
     group.add_argument("--restart-after-silence", action="store_true", help="Restart the recording after a transcription triggered by a silence")
 
     parser.add_argument("--download-folder-vosk", help="Folder to store Vosk models.")
@@ -249,7 +251,15 @@ def create_app(micro, transcriber, **kwargs):
         image_recording = Image.alpha_composite(image_recording.convert("RGBA"), image_writing.convert("RGBA"))
 
     def update_icon(icon, force=False):
-        if transcriber.recording:
+        if transcriber.recording and transcriber.waiting:
+            # this is the situation with the whisper backend when the microphone is recording
+            # but we wait for the speaker to speak (silence)
+            if force or getattr(icon, "_icon_label", None) != None:
+                icon.icon = image
+                icon._icon_label = None
+                icon.update_menu()
+
+        elif transcriber.recording:
             if force or getattr(icon, "_icon_label", None) != "recording":
                 icon.icon = image_recording
                 icon._icon_label = "recording"
@@ -361,6 +371,7 @@ def main(args=None):
                 if transcriber.backend == "whisper":
                     print(f"{colored('[t]', 'light_yellow')} change duration (currently {colored(transcriber.timeout, 'light_blue')} s)")
                     print(f"{colored('[b]', 'light_yellow')} change silence (currently {colored(transcriber.silence_duration, 'light_blue')} s)")
+                    print(f"{colored('[db]', 'light_yellow')} change backround noise (currently {colored(transcriber.silence_thresh, 'light_blue')} db)")
                     print(f"{colored('[a]', 'light_yellow')} auto-restart after silence is {colored(transcriber.restart_after_silence, 'light_blue')} toggle?")
                 exclude_flags = ["keyboard", "clipboard", "app", "prompt", "restart_after_silence"]
                 display_flags = [a.dest for a in parser._actions if a.help != argparse.SUPPRESS]
@@ -401,9 +412,9 @@ def main(args=None):
             if key == "t":
                 ans = input(f"Enter new duration in seconds (current: {transcriber.timeout}): ")
                 try:
-                    o.duration = transcriber.timeout = int(ans)
+                    o.duration = transcriber.timeout = float(ans)
                 except:
-                    print("Invalid duration. Must be an integer.")
+                    print("Invalid duration. Must be a float.")
                 continue
             if key == "latency":
                 ans = input(f"Enter new keyboard latency in seconds (current: {o.latency}): ")
@@ -415,9 +426,16 @@ def main(args=None):
             if key == "b":
                 ans = input(f"Enter new silence break duration in seconds (current: {transcriber.silence_duration}): ")
                 try:
-                    o.silence = transcriber.silence_duration = int(ans)
+                    o.silence = transcriber.silence_duration = float(ans)
+                except:
+                    print("Invalid duration. Must be a float.")
+                continue
+            if key == "db":
+                ans = input(f"Enter new background noise threshold to detect silence (current: {transcriber.silence_thresh}): ")
+                try:
+                    o.silence_db = transcriber.silence_thresh = float(ans)
                 except:
-                    print("Invalid duration. Must be an integer.")
+                    print("Invalid duration. Must be a float.")
                 continue
             if key:
                 if hasattr(o, key) and isinstance(getattr(o, key), bool):

{scribe_cli-0.7.10 → scribe_cli-0.7.11}/scribe/models.py

@@ -34,6 +34,7 @@ class AbstractTranscriber:
         self.restart_after_silence = restart_after_silence
         self.recording = False
         self.busy = False
+        self.waiting = False
         self.reset()
 
     def get_elapsed(self):
@@ -52,7 +53,6 @@ class AbstractTranscriber:
     def reset(self):
         self.audio_buffer = b''
         self.start_time = time.time()
-        self.last_sound_time = time.time()
 
     def start_recording(self, microphone,
                         start_message="Recording... Press Ctrl+C to stop.",
@@ -60,7 +60,13 @@ class AbstractTranscriber:
 
         self.reset()
         self.recording = True
+        self.waiting = True
         self.busy = True
+        if self.silence_duration is not None:
+            self.last_sound_time = time.time() - self.silence_duration
+        else:
+            self.last_sound_time = time.time()
+        previous_waiting = self.waiting
 
         try:
 
@@ -75,19 +81,31 @@ class AbstractTranscriber:
                         if is_silent(data, self.silence_thresh):
                             silence_duration = time.time() - self.last_sound_time
 
-                            if self.silence_duration is not None and silence_duration >= self.silence_duration and len(self.audio_buffer) > 0:
+                            previous_waiting = self.waiting
+                            self.waiting = self.silence_duration is not None and silence_duration >= self.silence_duration
+
+                            if self.waiting and len(self.audio_buffer) > 0:
                                 if self.restart_after_silence:
+                                    self.recording = False # for the system tray icon
                                     result = self.finalize()
                                     microphone.q.queue.clear()
                                     self.reset()
                                     yield result
+                                    self.recording = True # for the system tray icon
                                 else:
                                     raise StopRecording("Silence detected: {:.2f} seconds".format(silence_duration))
 
                         else:
                             self.last_sound_time = time.time()
+                            self.waiting = False
 
-                        yield self.transcribe_realtime_audio(data)
+                        # don't accumulate very long silences
+                        if not self.waiting:
+                            yield self.transcribe_realtime_audio(data)
+
+                        else:
+                            if not previous_waiting:
+                                print("Silence detected...waiting for more audio")
 
                         if self.is_overtime():
                             raise StopRecording("Overtime: {:.2f} seconds".format(self.get_elapsed()))
@@ -98,6 +116,7 @@ class AbstractTranscriber:
             pass
 
         finally:
+            self.waiting = False
             self.recording = False
             result = self.finalize()
             microphone.q.queue.clear()

{scribe_cli-0.7.10 → scribe_cli-0.7.11/scribe_cli.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.2
 Name: scribe-cli
-Version: 0.7.10
+Version: 0.7.11
 Summary: scribe is a local speech recognition tool that provides real-time transcription using vosk and whisper AI, with the goal of serving as a virtual keyboard on a computer
 Author-email: Mahé Perrette <mahe.perrette@gmail.com>
 License: MIT License
@@ -104,7 +104,7 @@ pip install -e .[all]
 You can leave the optional dependencies (leave out `[all]`) but must install at least one of `vosk` or `openai-whisper` packages (see Usage below).
 
 The `vosk` language models will download on-the-fly.
-The default download folder is `$XDG_CACHE_HOME/{backend}` where `$XDG_CACHE_HOME` defaults to `$HOME/.cache` (note for the `whisker` backend
+The default download folder is `$XDG_CACHE_HOME/{backend}` where `$XDG_CACHE_HOME` defaults to `$HOME/.cache` (note for the `whisper` backend
 the default is left to the `openai-whisper` package and might change in the future).
 
 
@@ -122,8 +122,8 @@ You can interrupt the recording via Ctrl + C and start again or change model. Th
 
 The default (`whisper`) is excellent at transcribing a full-length audio sequences in [many languages](https://github.com/openai/whisper?tab=readme-ov-file#available-models-and-languages). It is really impressive,
 but it cannot do real-time, and depending on the model can have relatively long execution time, especially with the `turbo` model (at least on my laptop with CPU only). The `small` model is also excellent and runs much faster. It is selected as default in `scribe` for that reason.
-With the `whisker` model you need to stop the registration manually before the transcription occurs (Ctrl + C), though
-there is a maximum duration after which it will stop by itself, which is setup to 60s by default (unless `--duration` is set to something else).
+With the `whisper` model the registration stops after a 2-second silence is detected. You can also stop the registration manually before the transcription occurs (Ctrl + C or Stop in the `--app` mode).
+By default, the recording will only last 120 seconds. You can fine-tune this behaviour via the `--silence`, `--duration` and `--restart-after-silence` parameters.
 
 The `vosk` backend is much faster and very good at doing real-time transcription for one language, but tended to make more mistakes in my tests and it does not do punctuation.
 It becomes really powerful when used for longer or interactive typing session with the [keyboard](#virtual-keyboard-experimental) option, e.g. to make notes or chat with an AI.