fix: openai api transcriber (#652)

buzz/settings/settings.py

@@ -34,6 +34,8 @@ class Key(enum.Enum):
             "transcription-tasks-table/column-visibility"
         )
 
+        MAIN_WINDOW = "main-window"
+
     def set_value(self, key: Key, value: typing.Any) -> None:
         self.settings.setValue(key.value, value)
 

buzz/transcriber.py

@@ -3,6 +3,7 @@
 import enum
 import json
 import logging
+import math
 import multiprocessing
 import os
 import subprocess
@@ -286,56 +287,100 @@ def transcribe(self) -> List[Segment]:
             self.task,
         )
 
-        wav_file = tempfile.mktemp() + ".wav"
+        mp3_file = tempfile.mktemp() + ".mp3"
 
-        # fmt: off
-        cmd = [
-            "ffmpeg",
-            "-nostdin",
-            "-threads", "0",
-            "-i", self.file_path,
-            "-f", "s16le",
-            "-ac", "1",
-            "-acodec", "pcm_s16le",
-            "-ar", str(whisper.audio.SAMPLE_RATE),
-            wav_file,
-        ]
-        # fmt: on
+        cmd = ["ffmpeg", "-i", self.file_path, mp3_file]
 
         try:
             subprocess.run(cmd, capture_output=True, check=True)
         except subprocess.CalledProcessError as exc:
             logging.exception("")
             raise Exception(exc.stderr.decode("utf-8"))
 
-        # TODO: Check if file size is more than 25MB (2.5 minutes), then chunk
-        audio_file = open(wav_file, "rb")
+        # fmt: off
+        cmd = [
+            "ffprobe",
+            "-v", "error",
+            "-show_entries", "format=duration",
+            "-of", "default=noprint_wrappers=1:nokey=1",
+            mp3_file,
+        ]
+        # fmt: on
+        duration_secs = float(
+            subprocess.run(cmd, capture_output=True, check=True).stdout.decode("utf-8")
+        )
+
+        total_size = os.path.getsize(mp3_file)
+        max_chunk_size = 25 * 1024 * 1024
+
         openai.api_key = (
             self.transcription_task.transcription_options.openai_access_token
         )
-        language = self.transcription_task.transcription_options.language
-        response_format = "verbose_json"
-        if self.transcription_task.transcription_options.task == Task.TRANSLATE:
-            transcript = openai.Audio.translate(
-                "whisper-1",
-                audio_file,
-                response_format=response_format,
-                language=language,
-            )
-        else:
-            transcript = openai.Audio.transcribe(
-                "whisper-1",
-                audio_file,
-                response_format=response_format,
-                language=language,
+
+        self.progress.emit((0, 100))
+
+        if total_size < max_chunk_size:
+            return self.get_segments_for_file(mp3_file)
+
+        # If the file is larger than 25MB, split into chunks
+        # and transcribe each chunk separately
+        num_chunks = math.ceil(total_size / max_chunk_size)
+        chunk_duration = duration_secs / num_chunks
+
+        segments = []
+
+        for i in range(num_chunks):
+            chunk_start = i * chunk_duration
+            chunk_end = min((i + 1) * chunk_duration, duration_secs)
+
+            chunk_file = tempfile.mktemp() + ".mp3"
+
+            # fmt: off
+            cmd = [
+                "ffmpeg",
+                "-i", mp3_file,
+                "-ss", str(chunk_start),
+                "-to", str(chunk_end),
+                "-c", "copy",
+                chunk_file,
+            ]
+            # fmt: on
+            subprocess.run(cmd, capture_output=True, check=True)
+            logging.debug('Created chunk file "%s"', chunk_file)
+
+            segments.extend(
+                self.get_segments_for_file(
+                    chunk_file, offset_ms=int(chunk_start * 1000)
+                )
             )
+            os.remove(chunk_file)
+            self.progress.emit((i + 1, num_chunks))
 
-        segments = [
-            Segment(segment["start"] * 1000, segment["end"] * 1000, segment["text"])
-            for segment in transcript["segments"]
-        ]
         return segments
 
+    def get_segments_for_file(self, file: str, offset_ms: int = 0):
+        with open(file, "rb") as audio_file:
+            kwargs = {
+                "model": "whisper-1",
+                "file": audio_file,
+                "response_format": "verbose_json",
+                "language": self.transcription_task.transcription_options.language,
+            }
+            transcript = (
+                openai.Audio.translate(**kwargs)
+                if self.transcription_task.transcription_options.task == Task.TRANSLATE
+                else openai.Audio.transcribe(**kwargs)
+            )
+
+            return [
+                Segment(
+                    int(segment["start"] * 1000 + offset_ms),
+                    int(segment["end"] * 1000 + offset_ms),
+                    segment["text"],
+                )
+                for segment in transcript["segments"]
+            ]
+
     def stop(self):
         pass
 

buzz/widgets/main_window.py

@@ -111,6 +111,8 @@ def __init__(self, tasks_cache=TasksCache()):
 
         self.load_tasks_from_cache()
 
+        self.load_geometry()
+
     def dragEnterEvent(self, event):
         # Accept file drag events
         if event.mimeData().hasUrls():
@@ -314,10 +316,27 @@ def on_shortcuts_changed(self, shortcuts: dict):
         self.toolbar.set_shortcuts(shortcuts=self.shortcuts)
         self.shortcut_settings.save(shortcuts=self.shortcuts)
 
+    def resizeEvent(self, event):
+        self.save_geometry()
+
     def closeEvent(self, event: QtGui.QCloseEvent) -> None:
+        self.save_geometry()
+
         self.transcriber_worker.stop()
         self.transcriber_thread.quit()
         self.transcriber_thread.wait()
         self.save_tasks_to_cache()
         self.shortcut_settings.save(shortcuts=self.shortcuts)
         super().closeEvent(event)
+
+    def save_geometry(self):
+        self.settings.begin_group(Settings.Key.MAIN_WINDOW)
+        self.settings.settings.setValue("geometry", self.saveGeometry())
+        self.settings.end_group()
+
+    def load_geometry(self):
+        self.settings.begin_group(Settings.Key.MAIN_WINDOW)
+        geometry = self.settings.settings.value("geometry")
+        if geometry is not None:
+            self.restoreGeometry(geometry)
+        self.settings.end_group()

buzz/widgets/transcriber/file_transcriber_widget.py

@@ -145,6 +145,8 @@ def __init__(
         self.setLayout(layout)
         self.setFixedSize(self.sizeHint())
 
+        self.reset_transcriber_controls()
+
     def get_on_checkbox_state_changed_callback(self, output_format: OutputFormat):
         def on_checkbox_state_changed(state: int):
             if state == Qt.CheckState.Checked.value:
@@ -158,11 +160,6 @@ def on_transcription_options_changed(
         self, transcription_options: TranscriptionOptions
     ):
         self.transcription_options = transcription_options
-        self.word_level_timings_checkbox.setDisabled(
-            self.transcription_options.model.model_type == ModelType.HUGGING_FACE
-            or self.transcription_options.model.model_type
-            == ModelType.OPEN_AI_WHISPER_API
-        )
         if self.transcription_options.openai_access_token != "":
             self.openai_access_token_changed.emit(
                 self.transcription_options.openai_access_token
@@ -213,6 +210,11 @@ def on_download_model_error(self, error: str):
 
     def reset_transcriber_controls(self):
         self.run_button.setDisabled(False)
+        self.word_level_timings_checkbox.setDisabled(
+            self.transcription_options.model.model_type == ModelType.HUGGING_FACE
+            or self.transcription_options.model.model_type
+            == ModelType.OPEN_AI_WHISPER_API
+        )
 
     def on_cancel_model_progress_dialog(self):
         if self.model_loader is not None:

buzz/widgets/transcription_tasks_table_widget.py

@@ -24,7 +24,7 @@ class TableColDef:
     id: str
     header: str
     column_index: int
-    value_getter: Callable[..., str]
+    value_getter: Callable[[FileTranscriptionTask], str]
     width: Optional[int] = None
     hidden: bool = False
     hidden_toggleable: bool = True
@@ -37,6 +37,8 @@ class Column(enum.Enum):
         MODEL = auto()
         TASK = auto()
         STATUS = auto()
+        DATE_ADDED = auto()
+        DATE_COMPLETED = auto()
 
     return_clicked = pyqtSignal()
 
@@ -78,7 +80,7 @@ def __init__(self, parent: Optional[QWidget] = None):
                 header=_("Task"),
                 column_index=self.Column.TASK.value,
                 value_getter=lambda task: self.get_task_label(task),
-                width=180,
+                width=120,
                 hidden=True,
             ),
             TableColDef(
@@ -89,6 +91,28 @@ def __init__(self, parent: Optional[QWidget] = None):
                 width=180,
                 hidden_toggleable=False,
             ),
+            TableColDef(
+                id="date_added",
+                header=_("Date Added"),
+                column_index=self.Column.DATE_ADDED.value,
+                value_getter=lambda task: task.queued_at.strftime("%Y-%m-%d %H:%M:%S")
+                if task.queued_at is not None
+                else "",
+                width=180,
+                hidden=True,
+            ),
+            TableColDef(
+                id="date_completed",
+                header=_("Date Completed"),
+                column_index=self.Column.DATE_COMPLETED.value,
+                value_getter=lambda task: task.completed_at.strftime(
+                    "%Y-%m-%d %H:%M:%S"
+                )
+                if task.completed_at is not None
+                else "",
+                width=180,
+                hidden=True,
+            ),
         ]
 
         self.setColumnCount(len(self.column_definitions))
@@ -155,8 +179,9 @@ def upsert_task(self, task: FileTranscriptionTask):
                 item.setFlags(item.flags() & ~Qt.ItemFlag.ItemIsEditable)
                 self.setItem(row_index, definition.column_index, item)
         else:
-            status_widget = self.item(task_row_index, self.Column.STATUS.value)
-            status_widget.setText(task.status_text())
+            for definition in self.column_definitions:
+                item = self.item(task_row_index, definition.column_index)
+                item.setText(definition.value_getter(task))
 
     @staticmethod
     def get_task_label(task: FileTranscriptionTask) -> str:

buzz/widgets/transcription_viewer/transcription_viewer_widget.py

@@ -62,6 +62,7 @@ def set_segment_text(self, text: str):
         self.task_changed.emit()
 
 
+# TODO: Fix player duration and add spacer below
 class TranscriptionViewerWidget(QWidget):
     transcription_task: FileTranscriptionTask
     task_changed = pyqtSignal()

tests/transcriber_test.py

@@ -26,6 +26,7 @@
     whisper_cpp_params,
     write_output,
     TranscriptionOptions,
+    OpenAIWhisperAPIFileTranscriber,
 )
 from buzz.recording_transcriber import RecordingTranscriber
 from tests.mock_sounddevice import MockInputStream
@@ -70,6 +71,37 @@ def test_should_transcribe(self, qtbot):
         assert "Bienvenue dans Passe" in text
 
 
+class TestOpenAIWhisperAPIFileTranscriber:
+    def test_transcribe(self):
+        file_path = "testdata/whisper-french.mp3"
+        transcriber = OpenAIWhisperAPIFileTranscriber(
+            task=FileTranscriptionTask(
+                file_path=file_path,
+                transcription_options=(
+                    TranscriptionOptions(
+                        openai_access_token=os.getenv("OPENAI_ACCESS_TOKEN"),
+                    )
+                ),
+                file_transcription_options=(
+                    FileTranscriptionOptions(file_paths=[file_path])
+                ),
+                model_path="",
+            )
+        )
+        mock_completed = Mock()
+        transcriber.completed.connect(mock_completed)
+        mock_openai_result = {"segments": [{"start": 0, "end": 6.56, "text": "Hello"}]}
+        with patch("openai.Audio.transcribe", return_value=mock_openai_result):
+            transcriber.run()
+
+        called_segments = mock_completed.call_args[0][0]
+
+        assert len(called_segments) == 1
+        assert called_segments[0].start == 0
+        assert called_segments[0].end == 6560
+        assert called_segments[0].text == "Hello"
+
+
 class TestWhisperCppFileTranscriber:
     @pytest.mark.parametrize(
         "word_level_timings,expected_segments",