neural-extractor-node 0.1.0__py3-none-any.whl

neural_extractor_node/__init__.py

@@ -0,0 +1 @@
+__version__ = "0.1.0"

neural_extractor_node/transcribe.py

@@ -0,0 +1,134 @@
+# /// script
+# # STRICT LOCK: Capped at 3.12. Python 3.13+ (PEP 667) breaks Whisper's setup.py read_version() via locals() KeyError.
+# requires-python = ">=3.11, <=3.12"
+# dependencies = [
+#     "openai-whisper==20240930",
+#     "torch==2.12.0",
+# # Whisper doesn't actually need torchaudio, because torchaudio==2.12.0 wasn't released, we're just dropping it out.
+# ]
+#
+# [tool.uv.sources]
+# # HARDWARE BRIDGE: Routes the engine downloads to the CUDA 13.0 binaries (Blackwell sm_120 support)
+# torch = { index = "pytorch-cu130" }
+#
+# [[tool.uv.index]]
+# name = "pytorch-cu130"
+# url = "https://download.pytorch.org/whl/cu130"
+# explicit = true
+#
+# [tool.uv.extra-build-dependencies]
+# # STRICT LOCK: OpenAI's setup.py still relies on pkg_resources, requiring legacy setuptools to build
+# openai-whisper = ["setuptools<81"]
+# ///
+
+import whisper
+import glob
+import os
+import re
+import torch
+
+
+# -------- CONFIG & HARDWARE --------
+MODEL_NAME = "medium"   # "small" = faster, "medium" = more accurate
+LANGUAGE = "ru" #"en"
+
+# --- HYBRID PATH DISCOVERY ---
+def get_data_dir():
+    # Priority 1: 'data' (Docker volume mount)
+    # Priority 2: '.' (Native: User is running from inside the neural_extractor_node folder)
+    # Priority 3: 'src/neural_extractor_node' (Native: User is running from the project root)
+    for candidate in ["data", ".", "src/neural_extractor_node"]:
+        if glob.glob(os.path.join(candidate, "*part*.wav")):
+            return candidate
+    return "." # Fallback if nothing is found
+
+BASE_DIR = get_data_dir()
+
+PATTERN = os.path.join(BASE_DIR, "*part*.wav")
+OUTPUT_FILE = os.path.join(BASE_DIR, "full_transcript.txt")
+# -----------------------------------
+
+
+# Global Hardware Detection
+DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
+# -----------------------------------
+
+def extract_part_number(filename):
+    match = re.search(r'part(\d+)', filename)
+    return int(match.group(1)) if match else 0
+
+def main():
+    print(f"\n[SYSTEM] Target Hardware: {DEVICE.upper()}")
+
+    if DEVICE == "cuda":
+        gpu_name = torch.cuda.get_device_name(0)
+        print(f"[SYSTEM] GPU Detected: {gpu_name}")
+
+        # Hardware optimization for RTX 30/40 series (Ampere/Ada architectures)
+        # Allows PyTorch to use TensorFloat32 for massive matrix math speedups
+        torch.backends.cuda.matmul.allow_tf32 = True
+
+    print(f"\nLoading '{MODEL_NAME}' model into VRAM...")
+    # Explicitly map model into GPU memory
+    model = whisper.load_model(MODEL_NAME, device=DEVICE)
+
+    files = glob.glob(PATTERN)
+    files = sorted(files, key=extract_part_number)
+
+    print("\nMatched files:")
+    for f in files:
+        print(f)
+
+    if not files:
+        print("\nNo files found. Check filename pattern.")
+        return
+
+    print(f"\nFound {len(files)} files\n")
+
+    with open(OUTPUT_FILE, "w", encoding="utf-8") as out:
+        for i, file in enumerate(files, 1):
+            print(f"\n[{i}/{len(files)}] Processing: {file}")
+
+            try:
+                # --- INFERENCE EXECUTION ---
+                result = model.transcribe(
+                    file,
+                    language=LANGUAGE,
+                    # Dynamic fp16 toggling:
+                    # fp16=True (CUDA) drastically speeds up inference & cuts VRAM usage by 50%
+                    # fp16=False (CPU) prevents float16 warnings/crashes when running on base CPU
+                    fp16=(DEVICE == "cuda"),
+                    verbose=True
+                )
+
+                # Always build from segments
+                text = " ".join(seg["text"].strip() for seg in result["segments"])
+
+                print(f"Finished: {file} | chars: {len(text)}")
+
+                out.write(f"\n--- {os.path.basename(file)} ---\n")
+                out.write(text + "\n")
+                out.flush()
+
+            except Exception as e:
+                print(f"Error processing {file}: {e}")
+                out.write(f"\n--- ERROR in {file} ---\n{e}\n")
+                out.flush()
+
+    print(f"\nDone. Transcript saved to: {OUTPUT_FILE}")
+
+if __name__ == "__main__":
+    main()
+
+# --- EXECUTION INSTRUCTIONS ---
+
+# Native Bare Metal Ubuntu (Ensure NVIDIA Proprietary Drivers + CUDA Toolkit are installed)
+# sudo apt update && sudo apt install -y ffmpeg
+# curl -LsSf https://astral.sh/uv/install.sh | sh
+# uv run python "$(pwd)/src/neural_extractor_node/transcribe.py"
+# or inside "src/neural_extractor_node"
+# uv run python "transcribe.py"
+
+# Docker Execution (If running this inside the container built previously):
+# MUST pass the GPU flag to Docker to expose the hardware bridge:
+# docker run -it --gpus all -v "$(pwd)/src/neural_extractor_node:/app/data" neural-extractor-node-i

neural_extractor_node-0.1.0.dist-info/METADATA

@@ -0,0 +1,7 @@
+Metadata-Version: 2.4
+Name: neural-extractor-node
+Version: 0.1.0
+Summary: Deterministic Whisper Pipeline
+Requires-Python: <=3.12,>=3.11
+Requires-Dist: openai-whisper==20240930
+Requires-Dist: torch==2.12.0

neural_extractor_node-0.1.0.dist-info/RECORD

@@ -0,0 +1,6 @@
+neural_extractor_node/__init__.py,sha256=kUR5RAFc7HCeiqdlX36dZOHkUI5wI6V_43RpEcD8b-0,22
+neural_extractor_node/transcribe.py,sha256=w0Yu5BJd3Hsn-v4EcK2x7T2KHCKiGqDO2een2jZKuHo,4680
+neural_extractor_node-0.1.0.dist-info/METADATA,sha256=G_7DMqrFb9d2Wop7Pqtcfd-bcvEFYp3qZItGYSnd7Io,205
+neural_extractor_node-0.1.0.dist-info/WHEEL,sha256=aeYiig01lYGDzBgS8HxWXOg3uV61G9ijOsup-k9o1sk,91
+neural_extractor_node-0.1.0.dist-info/top_level.txt,sha256=xKSWCynWxp-nrCnmmuKcz9QF6vEIMk3sYGNPWe9kKVA,22
+neural_extractor_node-0.1.0.dist-info/RECORD,,

neural_extractor_node-0.1.0.dist-info/WHEEL

@@ -0,0 +1,5 @@
+Wheel-Version: 1.0
+Generator: setuptools (82.0.1)
+Root-Is-Purelib: true
+Tag: py3-none-any
+

neural_extractor_node-0.1.0.dist-info/top_level.txt

@@ -0,0 +1 @@
+neural_extractor_node