Upload folder using huggingface_hub

Dockerfile

@@ -4,7 +4,6 @@ RUN apt-get update && \
     apt-get install -y --no-install-recommends ffmpeg && \
     rm -rf /var/lib/apt/lists/*
 
-# Create non-root user (HF Spaces requirement)
 RUN useradd -m -u 1000 user
 USER user
 ENV PATH="/home/user/.local/bin:$PATH"
@@ -13,9 +12,11 @@ WORKDIR /home/user/app
 COPY --chown=user:user requirements.txt .
 RUN pip install --no-cache-dir --user -r requirements.txt
 
+# Install faster-whisper separately (lighter than full whisperx)
+RUN pip install --no-cache-dir --user faster-whisper || echo "faster-whisper install skipped"
+
 COPY --chown=user:user . .
 RUN mkdir -p temp_jobs static
 
 EXPOSE 7860
-
 CMD ["python", "main.py"]

config.py

@@ -26,7 +26,7 @@ AUDIO_CHANNELS = 1
 
 # === TRANSCRIPTION: HuggingFace Free GPU API ===
 # This uses HF's free Inference API with GPU — no local model needed!
-HF_API_URL = "https://api-inference.huggingface.co/models/openai/whisper-large-v3"
+HF_API_URL = "https://api-inference.huggingface.co/models/openai/whisper-small"
 HF_TOKEN = os.getenv("HF_TOKEN", "")  # Optional: set for higher rate limits
 # Fallback to local faster-whisper if HF API fails
 WHISPER_MODEL_SIZE = os.getenv("WHISPER_MODEL", "tiny")

services/transcriber.py

@@ -26,27 +26,31 @@ def transcribe_audio(
     device: str = "cpu",
     progress_callback=None,
 ) -> List[Dict]:
-    """Transcribe audio — tries HF GPU API first, falls back to local."""
+    """Transcribe audio. Chain: HF API (GPU) → local faster-whisper → error."""
 
-    # Try HuggingFace API first (free GPU!)
+    # Try local faster-whisper FIRST (more reliable on Spaces)
     try:
-        logger.info("Attempting HuggingFace GPU transcription (whisper-large-v3)...")
+        import faster_whisper
+        logger.info("Using local faster-whisper (tiny model, CPU)...")
+        return _transcribe_local(audio_path, output_dir, source_language, progress_callback)
+    except ImportError:
+        logger.info("faster-whisper not available, trying HF API...")
+    except Exception as e:
+        logger.warning(f"Local transcription failed: {e}. Trying HF API...")
+
+    # Fallback: HuggingFace Inference API (free GPU)
+    try:
+        logger.info("Attempting HuggingFace API transcription (whisper-small)...")
         segments = _transcribe_hf_api(audio_path, output_dir, source_language, progress_callback)
         if segments and len(segments) > 0:
             logger.info(f"HF API transcription success: {len(segments)} segments")
             return segments
     except Exception as e:
-        logger.warning(f"HF API failed: {e}. Falling back to local.")
+        logger.error(f"HF API also failed: {e}")
 
-    # Fallback: local faster-whisper (if installed)
-    try:
-        logger.info("Using local faster-whisper fallback...")
-        return _transcribe_local(audio_path, output_dir, source_language, progress_callback)
-    except Exception as e2:
-        logger.error(f"Local fallback also failed: {e2}")
-        raise RuntimeError(
-            "Transcription failed. HF API may be overloaded. "
-            "Try again in a few minutes or set HF_TOKEN env var for priority access."
+    raise RuntimeError(
+        "Transcription failed with all methods. "
+        "The audio file may be too large or the service is overloaded. Try again."
         )