Update app.py

app.py

@@ -45,13 +45,11 @@ def create_vtt(segments, file_path):
     with open(file_path, "w", encoding="utf-8") as f:
         f.write("WEBVTT\n\n")
         for i, segment in enumerate(segments):
-            start_seconds = segment.get('start', 0)
-            end_seconds = segment.get('end', 0)
-            start = str(datetime.timedelta(seconds=int(start_seconds)))
-            end = str(datetime.timedelta(seconds=int(end_seconds)))
+            start = str(datetime.timedelta(seconds=int(segment["start"])))
+            end = str(datetime.timedelta(seconds=int(segment["end"])))
             f.write(f"{i+1}\n")
             f.write(f"{start} --> {end}\n")
-            f.write(f"{segment.get('text', '').strip()}\n\n")
+            f.write(f"{segment['text'].strip()}\n\n")
 
 def create_docx(segments, file_path, with_timestamps):
     document = Document()
@@ -59,14 +57,11 @@ def create_docx(segments, file_path, with_timestamps):
 
     if with_timestamps:
         for segment in segments:
-            text = segment.get('text', '').strip()
-            start_seconds = segment.get('start', 0)
-            end_seconds = segment.get('end', 0)
-            start = str(datetime.timedelta(seconds=int(start_seconds)))
-            end = str(datetime.timedelta(seconds=int(end_seconds)))
-            document.add_paragraph(f"[{start} - {end}] {text}")
+            start = str(datetime.timedelta(seconds=int(segment["start"])))
+            end = str(datetime.timedelta(seconds=int(segment["end"])))
+            document.add_paragraph(f"[{start} - {end}] {segment['text'].strip()}")
     else:
-        full_text = " ".join([segment.get('text', '').strip() for segment in segments])
+        full_text = " ".join([segment['text'].strip() for segment in segments])
         document.add_paragraph(full_text)
     
     document.save(file_path)
@@ -84,21 +79,41 @@ def chunk_audio_numpy(file_path, chunk_length_ms=60000):
         chunks.append((samples, len(chunk) / 1000))  # (array, duration_sec)
     return chunks
 
-def transcribe_long_audio(pipe, file_path, chunk_length_ms=60000):
+def transcribe_long_audio(pipe, file_path, chunk_length_ms=60000, language=None):
+    """
+    Transcribe long audio in chunks.
+    language: optional, e.g., 'fr' for French, 'en' for English
+    """
     chunks = chunk_audio_numpy(file_path, chunk_length_ms)
     segments = []
     offset = 0
 
     for samples, duration_sec in chunks:
-        # Feed waveform directly; remove sampling_rate
-        raw_output = pipe(samples, return_timestamps=True)
-        for seg in raw_output.get("chunks", []):
-            seg['start'] += offset
-            seg['end'] += offset
-            segments.append(seg)
+        kwargs = {"return_timestamps": True}
+        if language:
+            kwargs["language"] = language
+        
+        raw_output = pipe(samples, **kwargs)
+
+        # Depending on HF version, segments may be in 'chunks' or 'segments'
+        if "chunks" in raw_output:
+            seg_list = raw_output["chunks"]
+        elif "segments" in raw_output:
+            seg_list = raw_output["segments"]
+        else:
+            # fallback: entire text as one segment
+            seg_list = [{"start": 0, "end": duration_sec, "text": raw_output.get("text", "")}]
+
+        # Make sure start/end exist
+        for seg in seg_list:
+            start = seg.get("start", 0)
+            end = seg.get("end", duration_sec)
+            text = seg.get("text", "")
+            segments.append({"start": start + offset, "end": end + offset, "text": text})
+
         offset += duration_sec
 
-    full_text = " ".join([seg['text'] for seg in segments])
+    full_text = " ".join([seg["text"] for seg in segments])
     return segments, full_text
 
 # -------------------------
@@ -114,8 +129,11 @@ def transcribe_and_export(audio_file, model_size, vtt_output, docx_timestamp_out
     pipe = get_model_pipeline(model_size, progress)
     progress(0.75, desc="🎤 Transcribing audio...")
 
+    # Determine language: French-specific model defaults to 'fr'
+    language = "fr" if "FR" in model_size else None
+
     # Transcribe using chunked NumPy arrays
-    segments, transcribed_text = transcribe_long_audio(pipe, audio_file)
+    segments, transcribed_text = transcribe_long_audio(pipe, audio_file, language=language)
 
     outputs = {}
     progress(0.85, desc="📝 Generating output files...")