Update app.py

app.py

@@ -53,7 +53,51 @@ def apply_speaker_renames_to_df(df, fileIndex, column="task"):
     df = df.copy()
     df[column] = df[column].apply(lambda s: get_display_name(s, fileIndex))
     return df
-    
+
+def extract_speaker_clip(audio_path, annotation, speaker, clip_duration=5):
+    """
+    Extract a clip of up to clip_duration seconds for the given speaker.
+    Prefers the first segment >= clip_duration seconds; falls back to the longest segment.
+    Returns bytes (WAV) or None on failure.
+    """
+    try:
+        import io
+        # Collect all segments for this speaker
+        speaker_segments = [
+            (seg.start, seg.end)
+            for seg, _, label in annotation.itertracks(yield_label=True)
+            if label == speaker
+        ]
+        if not speaker_segments:
+            return None
+
+        # Prefer first segment long enough for a full clip
+        chosen_start, chosen_end = None, None
+        for start, end in speaker_segments:
+            if (end - start) >= clip_duration:
+                chosen_start, chosen_end = start, start + clip_duration
+                break
+
+        # Fall back: use the longest segment available
+        if chosen_start is None:
+            longest = max(speaker_segments, key=lambda s: s[1] - s[0])
+            chosen_start, chosen_end = longest
+
+        # Load audio and slice the clip
+        waveform, sample_rate = torchaudio.load(audio_path)
+        start_frame = int(chosen_start * sample_rate)
+        end_frame = min(int(chosen_end * sample_rate), waveform.shape[-1])
+        clip = waveform[:, start_frame:end_frame]
+
+        # Encode as WAV in memory
+        buffer = io.BytesIO()
+        torchaudio.save(buffer, clip, sample_rate, format="wav")
+        buffer.seek(0)
+        return buffer.read()
+    except Exception as e:
+        print(f"extract_speaker_clip error for {speaker}: {e}")
+        return None
+
 @st.cache_data
 def convert_df(df):
    return df.to_csv(index=False).encode('utf-8')
@@ -115,23 +159,7 @@ def processFile(filePath):
     totalTimeInSeconds = int(waveform_gain_adjusted.shape[-1]/sampleRate)
     print("Time in seconds calculated")
     return annotations, totalTimeInSeconds
-    
-@st.cache_data
-def extract_speaker_clip(file_path, speaker, _annotation, clip_duration=5.0):
-    waveform, sample_rate = torchaudio.load(file_path)
-    for segment, _, label in _annotation.itertracks(yield_label=True):
-        if label == speaker and segment.duration >= 1.0:
-            start_sample = int(segment.start * sample_rate)
-            end_sample = int(min(segment.start + clip_duration, segment.end) * sample_rate)
-            clip = waveform[:, start_sample:end_sample]
-            # Convert to wav bytes in memory
-            import io
-            buf = io.BytesIO()
-            torchaudio.save(buf, clip, sample_rate, format="wav")
-            buf.seek(0)
-            return buf.read()
-    return None
-    
+
 def addCategory():
     newCategory = st.session_state.categoryInput
     st.toast(f"Adding {newCategory}")
@@ -622,21 +650,39 @@ try:
        
         newCategory = st.sidebar.text_input('Add category', key='categoryInput',on_change=addCategory) 
 
+        # --- Speaker Sample Clips (audio files only) ---
+        _is_audio_file = currFile is not None and not any(
+            currFile.lower().endswith(ext) for ext in ('.rttm', '.txt', '.csv')
+        )
+        if _is_audio_file:
+            st.sidebar.divider()
+            st.sidebar.subheader("\U0001f3a7 Speaker Samples")
+            st.sidebar.caption(
+                "Listen to a 5-second clip for each speaker to help identify them."
+            )
+            _curr_audio_path = file_paths[currFileIndex] if currFileIndex < len(file_paths) else None
+            if _curr_audio_path and os.path.exists(_curr_audio_path):
+                for sp in speakerNames:
+                    _display = get_display_name(sp, currFileIndex)
+                    st.sidebar.markdown(f"**{_display}**")
+                    _clip_bytes = extract_speaker_clip(
+                        _curr_audio_path, currAnnotation, sp, clip_duration=5
+                    )
+                    if _clip_bytes:
+                        st.sidebar.audio(_clip_bytes, format="audio/wav")
+                    else:
+                        st.sidebar.caption("_(No audio segment found)_")
+            else:
+                st.sidebar.caption("_(Audio file not available for preview)_")
+
         st.sidebar.divider()
         st.sidebar.subheader("Rename Speakers")
-        st.sidebar.caption("Replace SPEAKER_## labels with real names. Listen to a sample clip to help identify each speaker.")
+        st.sidebar.caption("Replace SPEAKER_## labels with real names.")
         current_renames = st.session_state.speakerRenames[currFileIndex]
-        currFilePath = st.session_state.file_paths[currFileIndex]
         for sp in speakerNames:
             current_label = current_renames.get(sp, "")
-            st.sidebar.markdown(f"**{sp}**")
-            # Only show audio player for real audio files (not rttm/csv/txt)
-            if currFilePath.lower().endswith(('.wav', '.mp3', '.mp4')):
-                clip_bytes = extract_speaker_clip(currFilePath, sp, currAnnotation)
-                if clip_bytes:
-                    st.sidebar.audio(clip_bytes, format="audio/wav")
             new_name = st.sidebar.text_input(
-                f"Rename {sp}",
+                f"{sp}",
                 value=current_label,
                 placeholder=f"e.g. John",
                 key=f"rename_{currFileIndex}_{sp}"