audio and transcript

cht/window.py

@@ -10,9 +10,14 @@ gi.require_version("Adw", "1")
 gi.require_version("GdkPixbuf", "2.0")
 from gi.repository import Gtk, Gdk, Adw, GLib, Pango, GdkPixbuf
 
+from threading import Thread
+
 from cht.config import APP_NAME, SCENE_THRESHOLD
 from cht.ui.timeline import Timeline, TimelineControls
 from cht.ui.monitor import MonitorWidget
+from cht.ui.waveform import WaveformWidget
+from cht.audio.waveform import WaveformEngine
+from cht.transcriber.engine import TranscriberEngine
 from cht.stream.manager import StreamManager, list_sessions
 from cht.stream.tracker import RecordingTracker
 from cht.agent.runner import AgentRunner, ACTIONS, check_claude_cli
@@ -37,6 +42,8 @@ class ChtWindow(Adw.ApplicationWindow):
         # Timeline is the central state machine
         self._timeline = Timeline()
         self._agent = AgentRunner()
+        self._waveform_engine = WaveformEngine()
+        self._transcriber = TranscriberEngine()
 
         # Main layout
         self._main_paned = Gtk.Paned(orientation=Gtk.Orientation.HORIZONTAL)
@@ -165,6 +172,34 @@ class ChtWindow(Adw.ApplicationWindow):
         # Load existing frames into the strip
         self._load_existing_frames()
 
+        # Load existing transcript
+        transcript_index = self._stream_mgr.transcript_dir / "index.json"
+        if transcript_index.exists():
+            self._transcriber.load_index(transcript_index)
+            segs = self._transcriber.all_segments()
+            if segs:
+                self._append_transcript_segments(segs)
+                self._append_agent_output(f"  Loaded {len(segs)} transcript segments.\n")
+
+        # Compute waveform from existing recordings (background thread)
+        if segments:
+            from cht.stream import ffmpeg as ff
+
+            def _compute_waveform():
+                audio_dir = self._stream_mgr.audio_dir
+                audio_dir.mkdir(parents=True, exist_ok=True)
+                full_wav = audio_dir / "full.wav"
+                try:
+                    ff.extract_audio_chunk(segments[0], full_wav)
+                    self._waveform_engine.compute_full(full_wav)
+                    peaks = self._waveform_engine.peaks
+                    bucket_dur = self._waveform_engine.bucket_duration
+                    GLib.idle_add(self._waveform_widget.set_peaks, peaks.copy(), bucket_dur)
+                except Exception as e:
+                    log.error("Waveform computation failed: %s", e)
+
+            Thread(target=_compute_waveform, daemon=True, name="waveform_load").start()
+
         # Set up agent auth/model if not already done
         self._populate_model_dropdown()
 
@@ -197,6 +232,9 @@ class ChtWindow(Adw.ApplicationWindow):
         # Start scene detection
         self._stream_mgr.start_scene_detector(on_new_frames=self._on_new_scene_frames)
 
+        # Start audio extraction (waveform + transcription)
+        self._stream_mgr.start_audio_extractor(on_new_audio=self._on_new_audio)
+
         # Start polling for frame thumbnails
         GLib.timeout_add(1000, self._poll_frames)
 
@@ -237,6 +275,26 @@ class ChtWindow(Adw.ApplicationWindow):
         for f in frames:
             GLib.idle_add(self._timeline.add_scene_marker, f["timestamp"])
 
+    def _on_new_audio(self, wav_path, start_time, duration):
+        """Called from audio extractor thread with new WAV chunk."""
+        # Compute waveform peaks (fast, ~1ms)
+        self._waveform_engine.append_chunk(wav_path, start_time)
+        peaks = self._waveform_engine.peaks
+        bucket_dur = self._waveform_engine.bucket_duration
+        GLib.idle_add(self._waveform_widget.set_peaks, peaks.copy(), bucket_dur)
+
+        # Transcribe in separate thread (GPU-bound, ~1-2s per chunk)
+        def _transcribe():
+            new_segs = self._transcriber.transcribe_chunk(wav_path, time_offset=start_time)
+            if self._stream_mgr:
+                self._transcriber.save_index(
+                    self._stream_mgr.transcript_dir / "index.json"
+                )
+            if new_segs:
+                GLib.idle_add(self._append_transcript_segments, new_segs)
+
+        Thread(target=_transcribe, daemon=True, name="transcriber").start()
+
     def _check_recorder(self):
         """Watchdog: restart recorder if it died (sender disconnect, etc)."""
         if not self._streaming or not self._stream_mgr:
@@ -257,6 +315,10 @@ class ChtWindow(Adw.ApplicationWindow):
         log.info("Stopping stream...")
         self._timeline.reset()
         self._monitor.stop()
+        self._waveform_engine.reset()
+        self._waveform_widget.set_peaks(None, 0.05)
+        self._transcriber.reset()
+        self._transcript_view.get_buffer().set_text("")
         if self._tracker:
             self._tracker.stop()
             self._tracker = None
@@ -298,8 +360,10 @@ class ChtWindow(Adw.ApplicationWindow):
         stream_frame.set_child(self._monitor)
         top_paned.set_start_child(stream_frame)
 
-        self._waveform_area = self._build_placeholder("Waveform", height=250, width=200)
-        top_paned.set_end_child(self._waveform_area)
+        self._waveform_widget = WaveformWidget(self._timeline)
+        waveform_frame = Gtk.Frame()
+        waveform_frame.set_child(self._waveform_widget)
+        top_paned.set_end_child(waveform_frame)
         top_paned.set_position(650)
         right_box.append(top_paned)
 
@@ -819,6 +883,16 @@ class ChtWindow(Adw.ApplicationWindow):
         # Auto-scroll to bottom
         self._agent_output_view.scroll_to_iter(buf.get_end_iter(), 0, False, 0, 0)
 
+    def _append_transcript_segments(self, segments):
+        """Append transcription segments to the transcript panel."""
+        buf = self._transcript_view.get_buffer()
+        for seg in segments:
+            m1, s1 = divmod(int(seg.start), 60)
+            m2, s2 = divmod(int(seg.end), 60)
+            line = f"[{m1:02d}:{s1:02d}-{m2:02d}:{s2:02d}] {seg.id}  {seg.text}\n"
+            buf.insert(buf.get_end_iter(), line)
+        self._transcript_view.scroll_to_iter(buf.get_end_iter(), 0, False, 0, 0)
+
     # -- Frame thumbnails --
 
     def _load_existing_frames(self):