Update app.py

app.py

@@ -1,6 +1,5 @@
 import json
 import os
-import re
 import shutil
 import subprocess
 import tempfile
@@ -17,7 +16,6 @@ LANGUAGE_MAP = {
     "Bilingual": None,  # auto-detect
 }
 
-# Download the ONNX repo into the Space at startup.
 MODEL_DIR = snapshot_download(repo_id=REPO_ID)
 
 
@@ -54,43 +52,6 @@ def normalize_audio(input_path: str, progress: gr.Progress | None = None) -> str
     return str(out_path)
 
 
-def paragraphize_text(text: str, max_chars: int = 180, max_sentences: int = 3) -> str:
-    """Lightweight paragraphing that preserves the original wording."""
-    text = (text or "").strip()
-    if not text:
-        return ""
-
-    # Split on end-of-sentence punctuation for English and Chinese.
-    sentences = re.split(r"(?<=[\.\!\?\。\！？])\s+", text)
-    sentences = [s.strip() for s in sentences if s.strip()]
-
-    # Fallback: if no sentence punctuation exists, split by commas / Chinese commas
-    if len(sentences) <= 1:
-        chunks = re.split(r"(?<=[,，;；])\s*", text)
-        chunks = [c.strip() for c in chunks if c.strip()]
-        if len(chunks) > 1:
-            sentences = chunks
-
-    paragraphs = []
-    current = []
-    current_len = 0
-
-    for s in sentences:
-        proposed_len = current_len + (1 if current else 0) + len(s)
-        if current and (proposed_len > max_chars or len(current) >= max_sentences):
-            paragraphs.append(" ".join(current))
-            current = [s]
-            current_len = len(s)
-        else:
-            current.append(s)
-            current_len = proposed_len
-
-    if current:
-        paragraphs.append(" ".join(current))
-
-    return "\n\n".join(paragraphs)
-
-
 def run_onnx_asr(audio_path: str, mode: str, progress: gr.Progress | None = None) -> dict:
     if mode not in LANGUAGE_MAP:
         raise gr.Error("Invalid mode selected.")
@@ -122,7 +83,6 @@ def run_onnx_asr(audio_path: str, mode: str, progress: gr.Progress | None = None
         detail = stderr or stdout or "Unknown ASR error."
         raise gr.Error(detail[:1500]) from e
 
-    # Be resilient: find the last JSON object in stdout.
     output = (proc.stdout or "").strip().splitlines()
     parsed = None
     for line in reversed(output):
@@ -136,7 +96,6 @@ def run_onnx_asr(audio_path: str, mode: str, progress: gr.Progress | None = None
             continue
 
     if not isinstance(parsed, dict):
-        # Fallback: return raw text if the script prints plain text instead.
         return {
             "text": (proc.stdout or "").strip(),
             "language": None,
@@ -145,15 +104,15 @@ def run_onnx_asr(audio_path: str, mode: str, progress: gr.Progress | None = None
     return parsed
 
 
-def make_txt_file(text: str, original_audio_path: str, suffix: str) -> str:
+def make_txt_file(text: str, original_audio_path: str) -> str:
     out_dir = Path(tempfile.mkdtemp())
     stem = Path(original_audio_path).stem or "transcript"
-    out_path = out_dir / f"{stem}_{suffix}.txt"
+    out_path = out_dir / f"{stem}.txt"
     out_path.write_text(text, encoding="utf-8")
     return str(out_path)
 
 
-def transcribe(audio_file: str, mode: str, paragraphing: bool, progress=gr.Progress()):
+def transcribe(audio_file: str, mode: str, progress=gr.Progress()):
     if not audio_file:
         raise gr.Error("Please upload an audio file.")
 
@@ -164,14 +123,8 @@ def transcribe(audio_file: str, mode: str, paragraphing: bool, progress=gr.Progr
         normalized_path = normalize_audio(audio_file, progress=progress)
         result = run_onnx_asr(normalized_path, mode=mode, progress=progress)
 
-        raw_text = (result.get("text") or result.get("transcript") or "").strip()
-        if not raw_text:
-            raw_text = ""
-
-        final_text = paragraphize_text(raw_text) if paragraphing else raw_text
-
-        raw_txt = make_txt_file(raw_text, audio_file, "raw")
-        final_txt = make_txt_file(final_text, audio_file, "paragraphs" if paragraphing else "transcript")
+        text = (result.get("text") or result.get("transcript") or "").strip()
+        txt_file = make_txt_file(text, audio_file)
 
         detected_language = result.get("language") or result.get("detected_language")
         info = f"Mode: {mode}"
@@ -179,7 +132,7 @@ def transcribe(audio_file: str, mode: str, paragraphing: bool, progress=gr.Progr
             info += f"\nDetected language: {detected_language}"
 
         progress(1.0, desc="Done")
-        return raw_text, final_text, final_txt, info
+        return text, txt_file, info
 
     finally:
         if normalized_path and os.path.exists(normalized_path):
@@ -208,23 +161,16 @@ with gr.Blocks(title="Qwen3 ASR ONNX CPU") as demo:
             info="Bilingual means auto-detect.",
         )
 
-    paragraphing = gr.Checkbox(
-        value=True,
-        label="Auto paragraphing",
-        info="Preserves wording and only inserts paragraph breaks.",
-    )
-
     transcribe_btn = gr.Button("Transcribe")
 
-    raw_transcript = gr.Textbox(label="Raw transcript", lines=10)
-    formatted_transcript = gr.Textbox(label="Formatted transcript", lines=14)
+    transcript = gr.Textbox(label="Transcript", lines=14)
     download_file = gr.File(label="Download transcript")
     metadata = gr.Textbox(label="Info", lines=2, interactive=False)
 
     transcribe_btn.click(
         fn=transcribe,
-        inputs=[audio, mode, paragraphing],
-        outputs=[raw_transcript, formatted_transcript, download_file, metadata],
+        inputs=[audio, mode],
+        outputs=[transcript, download_file, metadata],
     )
 
 if __name__ == "__main__":