fix: restore FunClip transcription and multi-segment clipping

Use the Hugging Face Paraformer, VAD, and punctuation pipeline with explicit submodel hubs and timestamp generation. Also accept Gradio checkbox labels and interleave FFmpeg concat inputs for multiple clips.

app.py

@@ -3,16 +3,18 @@ import json
 import tempfile
 import subprocess
 import gradio as gr
-import numpy as np
-import torch
 
 from funasr import AutoModel
 
 model = AutoModel(
-    model="iic/speech_paraformer-large-vad-punc_asr_nat-zh-cn-16k-common-vocab8404-pytorch",
+    model="funasr/paraformer-zh",
     hub="hf",
-    model_hub="hf",
+    vad_model="funasr/fsmn-vad",
+    vad_kwargs={"hub": "hf", "max_single_segment_time": 60000},
+    punc_model="funasr/ct-punc",
+    punc_kwargs={"hub": "hf"},
     device="cpu",
+    disable_update=True,
 )
 
 
@@ -38,7 +40,12 @@ def transcribe_video(video_path, progress=gr.Progress()):
 
     progress(0.3, desc="Transcribing speech...")
     try:
-        res = model.generate(input=audio_path, batch_size_s=300)
+        res = model.generate(
+            input=audio_path,
+            batch_size_s=300,
+            pred_timestamp=True,
+            sentence_timestamp=True,
+        )
     except Exception as e:
         return f"Transcription error: {str(e)}", [], None
     finally:
@@ -75,7 +82,7 @@ def clip_video(video_path, sentences_json, selected_indices):
 
     sentences = json.loads(sentences_json)
 
-    indices = [int(i) for i in selected_indices]
+    indices = [int(str(i).split(":", 1)[0]) for i in selected_indices]
     if not indices:
         return None, "No segments selected."
 
@@ -103,9 +110,8 @@ def clip_video(video_path, sentences_json, selected_indices):
             f"[0:a]atrim=start={start:.3f}:end={end:.3f},asetpts=PTS-STARTPTS[a{i}];"
         )
 
-    concat_v = "".join(f"[v{i}]" for i in range(len(merged)))
-    concat_a = "".join(f"[a{i}]" for i in range(len(merged)))
-    filter_parts.append(f"{concat_v}{concat_a}concat=n={len(merged)}:v=1:a=1[outv][outa]")
+    concat_inputs = "".join(f"[v{i}][a{i}]" for i in range(len(merged)))
+    filter_parts.append(f"{concat_inputs}concat=n={len(merged)}:v=1:a=1[outv][outa]")
 
     filter_complex = "".join(filter_parts)