Update app.py

app.py

@@ -54,14 +54,18 @@ def apply_speaker_renames_to_df(df, fileIndex, column="task"):
     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):
+def extract_speaker_clip(annotation, speaker, waveform, sample_rate, clip_duration=5):
     """
-    Extract a clip of up to clip_duration seconds for the given speaker.
-    Returns (bytes, None) on success, or (None, error_string) on failure.
+    Slice a clip directly from the already-loaded waveform tensor.
+    Avoids torchaudio.load (broken in this env due to torchcodec/libnppicc).
+    Returns (wav_bytes, None) on success or (None, error_str) on failure.
     """
-    import io, traceback
+    import io, traceback as tb
     try:
-        # --- Step 1: collect segments for this speaker ---
+        if waveform is None or sample_rate is None:
+            return None, "No waveform stored (only audio uploads carry waveform data)"
+
+        # Collect this speaker's segments
         speaker_segments = []
         try:
             for seg, _, label in annotation.itertracks(yield_label=True):
@@ -71,9 +75,9 @@ def extract_speaker_clip(audio_path, annotation, speaker, clip_duration=5):
             return None, f"itertracks failed: {e}"
 
         if not speaker_segments:
-            return None, f"No segments found for speaker '{speaker}' in annotation"
+            return None, f"No segments found for speaker '{speaker}'"
 
-        # --- Step 2: pick best segment ---
+        # Prefer first segment >= clip_duration; else take longest
         chosen_start, chosen_end = None, None
         for start, end in speaker_segments:
             if (end - start) >= clip_duration:
@@ -83,45 +87,20 @@ def extract_speaker_clip(audio_path, annotation, speaker, clip_duration=5):
             longest = max(speaker_segments, key=lambda s: s[1] - s[0])
             chosen_start, chosen_end = longest
 
-        # --- Step 3: load audio with torchaudio, fallback to soundfile ---
-        waveform, sample_rate = None, None
-        load_errors = []
-        try:
-            waveform, sample_rate = torchaudio.load(audio_path)
-        except Exception as e:
-            load_errors.append(f"torchaudio.load: {e}")
-
-        if waveform is None:
-            try:
-                import soundfile as sf
-                import numpy as np
-                data, sample_rate = sf.read(audio_path, dtype='float32')
-                if data.ndim == 1:
-                    data = data[np.newaxis, :]
-                else:
-                    data = data.T
-                waveform = torch.from_numpy(data)
-            except Exception as e:
-                load_errors.append(f"soundfile: {e}")
-
-        if waveform is None:
-            return None, "Could not load audio: " + " | ".join(load_errors)
-
-        # --- Step 4: slice and encode ---
+        # Slice waveform
         start_frame = int(chosen_start * sample_rate)
-        end_frame = min(int(chosen_end * sample_rate), waveform.shape[-1])
+        end_frame   = min(int(chosen_end * sample_rate), waveform.shape[-1])
         clip = waveform[:, start_frame:end_frame]
 
-        buffer = io.BytesIO()
-        torchaudio.save(buffer, clip, sample_rate, format="wav")
-        buffer.seek(0)
-        return buffer.read(), None
+        # Encode to WAV bytes
+        buf = io.BytesIO()
+        torchaudio.save(buf, clip.cpu(), sample_rate, format="wav")
+        buf.seek(0)
+        return buf.read(), None
 
     except Exception as e:
-        msg = traceback.format_exc()
-        print(f"extract_speaker_clip unexpected error for {speaker}: {msg}")
+        print(f"extract_speaker_clip error for {speaker}: {tb.format_exc()}")
         return None, str(e)
-
 @st.cache_data
 def convert_df(df):
    return df.to_csv(index=False).encode('utf-8')
@@ -182,7 +161,7 @@ def processFile(filePath):
     print("Speakers Detected")
     totalTimeInSeconds = int(waveform_gain_adjusted.shape[-1]/sampleRate)
     print("Time in seconds calculated")
-    return annotations, totalTimeInSeconds
+    return annotations, totalTimeInSeconds, waveform_gain_adjusted, sampleRate
 
 def addCategory():
     newCategory = st.session_state.categoryInput
@@ -209,7 +188,7 @@ def updateCategoryOptions(resultIndex):
     #st.info(f"Updating result {resultIndex}")
     #st.info(f"In update: {st.session_state.categorySelect}")
     # Handle 
-    _, currAnnotation, _ = st.session_state.results[currFileIndex]
+    _r = st.session_state.results[currFileIndex]; currAnnotation = _r[0]
     speakerNames = currAnnotation.labels()
 
     # Handle speaker category sidebars
@@ -242,7 +221,7 @@ def analyze(inFileName):
             
             printV(f'In if',4)
             # Handle 
-            currAnnotation, currTotalTime = st.session_state.results[currFileIndex]
+            _r = st.session_state.results[currFileIndex]; currAnnotation, currTotalTime = _r[0], _r[1]
             speakerNames = currAnnotation.labels()
             printV(f'Loaded results',4)
             # Update other categories
@@ -567,9 +546,9 @@ else:
                         st.session_state.unusedSpeakers[i] = speakerNames
                 else:
                     with st.spinner(text=f'Processing File {i+1} of {totalFiles}'):
-                        annotations, totalSeconds = processFile(file_paths[i])
+                        annotations, totalSeconds, wf_stored, sr_stored = processFile(file_paths[i])
                         print(f"Finished processing {file_paths[i]}")
-                        st.session_state.results[i] = (annotations, totalSeconds)
+                        st.session_state.results[i] = (annotations, totalSeconds, wf_stored, sr_stored)
                         print("Results saved")
                         st.session_state.summaries[i] = {}
                         print("Summaries saved")
@@ -652,7 +631,10 @@ try:
         graphNames = ["Data","Voice Categories","Speaker Percentage","Speakers with Categories","Treemap","Timeline","Time Spoken"]
         dataTab, pie1, pie2, sunburst1, treemap1, timeline, bar1 = st.tabs(graphNames)
         # Handle 
-        currAnnotation, currTotalTime = st.session_state.results[currFileIndex]
+        _r = st.session_state.results[currFileIndex]
+        currAnnotation, currTotalTime = _r[0], _r[1]
+        currWaveform   = _r[2] if len(_r) > 2 else None
+        currSampleRate = _r[3] if len(_r) > 3 else None
         speakerNames = currAnnotation.labels()
     
         speakers_dataFrame = st.session_state.summaries[currFileIndex]["speakers_dataFrame"]
@@ -692,14 +674,14 @@ try:
                     _display = get_display_name(sp, currFileIndex)
                     st.sidebar.markdown(f"**{_display}**")
                     _clip_bytes, _clip_err = extract_speaker_clip(
-                        _curr_audio_path, currAnnotation, sp, clip_duration=5
+                        currAnnotation, sp, currWaveform, currSampleRate, clip_duration=5
                     )
                     if _clip_bytes:
                         st.sidebar.audio(_clip_bytes, format="audio/wav")
                     else:
                         st.sidebar.error(f"Clip error: {_clip_err}")
             else:
-                st.sidebar.warning(f"Audio file not found at path: {_curr_audio_path}")
+                st.sidebar.warning(f"Audio path not found: {_curr_audio_path}")
 
         st.sidebar.divider()
         st.sidebar.subheader("Rename Speakers")
@@ -1120,7 +1102,7 @@ if len(st.session_state.results) > 0:
                 }
                 allCategories = copy.deepcopy(st.session_state.categories)
                 for i in indices:
-                    currAnnotation, currTotalTime = st.session_state.results[i]
+                    _ri = st.session_state.results[i]; currAnnotation, currTotalTime = _ri[0], _ri[1]
                     categorySelections = st.session_state["categorySelect"][i]
                     catSummary,extraCats = su.calcCategories(currAnnotation,categorySelections)
                     st.session_state.summaries[i]["categories"] = (catSummary,extraCats)