Create app.py

app.py

@@ -0,0 +1,77 @@
+import os
+import tempfile
+import uuid
+from pathlib import Path
+import gradio as gr
+import ffmpeg
+from faster_whisper import WhisperModel
+
+
+# --- Helper functions ---
+def _format_timestamp(seconds: float) -> str:
+ms = int(round(seconds * 1000))
+hours = ms // 3600000
+ms_rem = ms % 3600000
+minutes = ms_rem // 60000
+ms_rem = ms_rem % 60000
+secs = ms_rem // 1000
+millis = ms_rem % 1000
+return f"{hours:02d}:{minutes:02d}:{secs:02d},{millis:03d}"
+
+
+
+
+def segments_to_srt(segments: list) -> str:
+lines = []
+for i, seg in enumerate(segments, start=1):
+start_ts = _format_timestamp(seg['start'])
+end_ts = _format_timestamp(seg['end'])
+text = seg['text'].replace('\n', ' ').strip()
+if not text:
+continue
+block = f"{i}\n{start_ts} --> {end_ts}\n{text}\n"
+lines.append(block)
+return "\n".join(lines)
+
+
+
+
+# --- Configuration ---
+MODEL_NAME = "Systran/faster-whisper-small" # small model for HF free CPU
+DEVICE = "cpu"
+OUTPUT_DIR = Path("outputs/subtitles")
+OUTPUT_DIR.mkdir(parents=True, exist_ok=True)
+
+
+print(f"Loading model {MODEL_NAME} on {DEVICE} ...")
+model = WhisperModel(MODEL_NAME, device=DEVICE)
+print("Model loaded.")
+
+
+
+
+def extract_audio(input_path: str, out_path: str):
+try:
+_ = (
+ffmpeg
+.input(input_path)
+.output(out_path, format='wav', acodec='pcm_s16le', ac=1, ar='16000')
+.overwrite_output()
+.run(quiet=True)
+)
+except ffmpeg.Error as e:
+stderr = getattr(e, 'stderr', None)
+msg = stderr.decode() if stderr else str(e)
+raise RuntimeError(f"ffmpeg error: {msg}")
+
+
+
+
+def transcribe_file_to_srt(file_obj, language: str = "en"):
+filename = getattr(file_obj, 'name', getattr(file_obj, 'filename', f"upload_{uuid.uuid4()}.bin"))
+input_filepath = Path(filename)
+
+
+tmp_dir = Path(tempfile.mkdtemp(prefix="subgen_"))
+saved_input = tmp_dir / (str(uuid.uuid4()) + input_filepath.suffix)
+demo.launch(share=True)