taco/microWakeWord-Trainer-Nvidia-Docker
1
0
Fork 0
mirror of https://github.com/TaterTotterson/microWakeWord-Trainer-Nvidia-Docker.git synced 2026-06-12 20:10:19 -06:00
Code Issues Packages Projects Releases Wiki Activity

Update advanced_training_notebook.ipynb

Browse Source
This commit is contained in:
MasterPhooey
2025-01-05 11:53:30 -06:00
committed by GitHub
parent 53fae46446
commit 65adcdbc8f
1 changed files with 28 additions and 19 deletions
Download Patch File Download Diff File Expand all files Collapse all files

47
advanced_training_notebook.ipynb
View File

@@ -136,7 +136,7 @@
"id": "YJRG4Qvo9nXG"
},
"outputs": [],
"source": [
"source": [
"# Downloads audio data for augmentation. This can be slow!\n",
"# Borrowed from openWakeWord's automatic_model_training.ipynb, accessed March 4, 2024\n",
"#\n",
@@ -174,8 +174,7 @@
"# Download and Process Audioset\n",
"# -----------------------------\n",
"audioset_dir = \"./audioset\"\n",
"if not os.path.exists(audioset_dir):\n",
" os.mkdir(audioset_dir)\n",
"os.makedirs(audioset_dir, exist_ok=True)\n",
"\n",
"# Full-scale dataset download links\n",
"dataset_links = [\n",
@@ -183,46 +182,56 @@
" for i in range(10)\n",
"]\n",
"\n",
"# Download and extract each dataset part\n",
"for link in dataset_links:\n",
" file_name = link.split(\"/\")[-1]\n",
" out_dir = os.path.join(audioset_dir, file_name)\n",
" if not os.path.exists(out_dir):\n",
" out_path = os.path.join(audioset_dir, file_name)\n",
" if not os.path.exists(out_path):\n",
" print(f\"Downloading {file_name}...\")\n",
" os.system(f\"wget --quiet -O {out_dir} {link}\")\n",
" os.system(f\"wget --quiet -O {out_path} {link}\")\n",
" print(f\"Extracting {file_name}...\")\n",
" os.system(f\"cd {audioset_dir} && tar -xf {file_name}\")\n",
" os.system(f\"tar -xf {out_path} -C {audioset_dir}\")\n",
"\n",
"# Directory for 16kHz WAV files\n",
"output_dir = \"./audioset_16k\"\n",
"if not os.path.exists(output_dir):\n",
" os.mkdir(output_dir)\n",
"os.makedirs(output_dir, exist_ok=True)\n",
"\n",
"# Save clips to 16-bit PCM wav files\n",
"audioset_files = list(Path(\"audioset/audio\").glob(\"**/*.flac\"))\n",
"# Collect all FLAC files for processing\n",
"audioset_files = list(Path(audioset_dir).glob(\"**/*.flac\"))\n",
"print(f\"Number of FLAC files found: {len(audioset_files)}\")\n",
"\n",
"if audioset_files:\n",
" audioset_dataset = Dataset.from_dict({\"audio\": [str(file) for file in audioset_files]})\n",
" audioset_dataset = audioset_dataset.cast_column(\"audio\", Audio(sampling_rate=16000))\n",
"\n",
" corrupted_files = []\n",
" print(\"Converting Audioset files to 16kHz WAV...\")\n",
" for row in tqdm(audioset_dataset):\n",
"\n",
" for row in tqdm(audioset_dataset, desc=\"Processing Audioset files\"):\n",
" try:\n",
" name = row[\"audio\"][\"path\"].split(\"/\")[-1].replace(\".flac\", \".wav\")\n",
" scipy.io.wavfile.write(\n",
" os.path.join(output_dir, name), \n",
" 16000, \n",
" (row[\"audio\"][\"array\"] * 32767).astype(np.int16)\n",
" )\n",
" # Define output file name and path\n",
" name = Path(row[\"audio\"][\"path\"]).stem + \".wav\"\n",
" output_path = Path(output_dir) / name\n",
" \n",
" # Read and convert audio\n",
" data = row[\"audio\"][\"array\"]\n",
" scipy.io.wavfile.write(output_path, 16000, (data * 32767).astype(np.int16))\n",
" \n",
" except Exception as e:\n",
" print(f\"Error converting {row['audio']['path']}: {e}\")\n",
" corrupted_files.append(row[\"audio\"][\"path\"])\n",
"\n",
" # Log corrupted files\n",
" if corrupted_files:\n",
" with open(\"audioset_corrupted_files.log\", \"w\") as log_file:\n",
" log_path = Path(output_dir) / \"audioset_corrupted_files.log\"\n",
" with open(log_path, \"w\") as log_file:\n",
" log_file.writelines(f\"{file}\\n\" for file in corrupted_files)\n",
" print(f\"Logged corrupted files to {log_path}\")\n",
"else:\n",
" print(\"No FLAC files found in Audioset.\")\n",
"\n",
"print(\"Audioset processing complete!\")\n",
"\n",
"# -----------------------------\n",
"# Download and Process FMA\n",
"# -----------------------------\n",