add dataset builder

README.md

@@ -4,4 +4,20 @@
 The [Northeastern Neo-Aramaic (NENA) Database Project](https://nena.ames.cam.ac.uk/) has been aggregating language documentation materials for the NENA dialects. These materials include descriptions, including [a description of the dialect of the Assyrian Christians of Urmi](https://drive.google.com/file/d/1k7QXjjxakQN87c0p-SAcUwnxY_JbrKj9/view?usp=drive_link). This description contains 300 pages (8 hours) of transcribed and translated oral literature. These oral literatures are [actively being parsed](https://github.com/mattynaz/nena-dataset-parsing) and uploaded to a database at [pocketbase.nenadb.dev](https://pocketbase.nenadb.dev/_). The platform [crowdsource.nenadb.dev](https://crowdsource.nenadb.dev/) allows the community to directly engage with these parsed examples and contribute their own voices to the database.
 
 ## Goal
-The goal is to publish this dataset to [HuggingFace](https://huggingface.co/). Mozilla's [Common Voice dataset](https://huggingface.co/datasets/mozilla-foundation/common_voice_13_0/tree/main) provides an example implementation of such a dataset.
+The goal is to publish this dataset to [HuggingFace](https://huggingface.co/). Mozilla's [Common Voice dataset](https://huggingface.co/datasets/mozilla-foundation/common_voice_13_0/tree/main) provides an example implementation of such a dataset. 
+
+## Development
+
+### Building the dataset
+
+Install the required packages.
+
+```
+pip install -r requirements.txt
+```
+
+Build the dataset.
+
+```
+python build.py --build
+```

build.py

@@ -0,0 +1,118 @@
+import argparse
+import csv
+import os
+import shutil
+import tarfile
+import tempfile
+from tqdm import tqdm
+
+from pydub import AudioSegment
+import requests
+from pocketbase import PocketBase
+
+parser = argparse.ArgumentParser(description="Command description.")
+
+pb = PocketBase('https://pocketbase.nenadb.dev/')
+
+def get_examples():
+    examples = pb.collection("examples").get_full_list(query_params={
+        "expand": "dialect",
+        "filter": "validated=true",
+    })
+
+    return examples
+
+def split_examples(examples, test_split=0.10, dev_split=0.10):
+    subsets = {}
+
+    for example in examples:
+        dialect = example.expand['dialect'].name.lower()
+        if not subsets.get(dialect):
+            subsets[dialect] = { 'all': [] }
+        subsets[dialect]['all'].append(example)
+
+    for subset in subsets.values():
+        for i, example in enumerate(subset['all']):
+            prog = i / len(subset['all'])
+
+            if prog < test_split:
+                split = 'test'
+            elif prog < dev_split + test_split:
+                split = 'dev'
+            else:
+                split = 'train'
+
+            if not subset.get(split):
+                subset[split] = []
+            subset[split].append(example)
+        
+        del subset['all']
+
+    return subsets
+
+def save_data(subsets):
+    total_examples = sum(
+        sum(len(split) for split in subset.values())
+        for subset in subsets.values()
+    )
+
+    with tqdm(total=total_examples) as pbar:
+        for dialect, subset in subsets.items():
+            for split, examples in subset.items():
+                audio_dir_path = os.path.join("audio", dialect, split)
+                os.makedirs(audio_dir_path, exist_ok=True)
+
+                transcripts = []
+                transcript_dir_path = os.path.join("transcript", dialect)
+                os.makedirs(transcript_dir_path, exist_ok=True)
+                
+                for example in examples:
+                    pbar.set_description(f"Downloading audios ({dialect} / {split})")
+                    pbar.update(1)
+                    audio_url = pb.get_file_url(example, example.speech, {})
+                    response = requests.get(audio_url)
+                    with tempfile.NamedTemporaryFile() as f:
+                        f.write(response.content)
+                        f.flush()
+                        audio = AudioSegment.from_file(f.name)
+                    audio = audio.set_frame_rate(48000)
+                    audio_file_name = f"nena_speech_{example.id}.mp3"
+                    audio_file_path = os.path.join(audio_dir_path, audio_file_name)
+                    audio.export(audio_file_path, format="mp3")
+                    
+                    transcripts.append({
+                        'age': example.age,
+                        'transcription': example.transcription,
+                        'translation': example.translation,
+                        'path': audio_file_name,
+                    })
+
+                pbar.set_description(f"Saving audios ({dialect}/{split})")
+                audio_tar_path = f"{audio_dir_path}.tar"
+                with tarfile.open(audio_tar_path, 'w') as tar:
+                    tar.add(audio_dir_path, arcname=os.path.basename(audio_dir_path))
+
+                pbar.set_description(f"Saving transcripts ({dialect} / {split})")
+                with open(os.path.join(transcript_dir_path, f"{split}.tsv"), 'w', newline='') as f:
+                    writer = csv.DictWriter(f, fieldnames=transcripts[0].keys(), delimiter='\t')
+                    writer.writeheader()
+                    writer.writerows(transcripts)
+
+                shutil.rmtree(audio_dir_path)
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(description="Generate text from prompt")
+
+    parser.add_argument(
+        "-b",
+        "--build",
+        action="store_true",
+        help="Download text prompts from GCS bucket",
+    )
+
+    args = parser.parse_args()
+
+    examples = get_examples()
+    subsets = split_examples(examples)
+    save_data(subsets)

main.ipynb

@@ -16,22 +16,81 @@
   },
   {
    "cell_type": "code",
-   "execution_count": 1,
+   "execution_count": 8,
    "metadata": {},
    "outputs": [],
    "source": [
     "from pocketbase import PocketBase\n",
     "\n",
-    "pb = PocketBase('https://pocketbase.nenadb.dev/')\n",
+    "def get_examples():\n",
+    "    pb = PocketBase('https://pocketbase.nenadb.dev/')\n",
     "\n",
-    "dialects = pb.collection(\"dialects\").get_full_list(query_params={\n",
-    "    \"sort\": \"name\",\n",
-    "})\n",
+    "    examples = pb.collection(\"examples\").get_full_list(query_params={\n",
+    "        \"expand\": \"dialect\",\n",
+    "        \"filter\": \"validated=true\",\n",
+    "    })\n",
     "\n",
-    "examples = pb.collection(\"examples\").get_full_list(query_params={\n",
-    "    \"expand\": \"dialect\",\n",
-    "    \"filter\": \"validated=true\",\n",
-    "})"
+    "    return examples"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 9,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "examples = get_examples()"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Bucket examples into subsets"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 10,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "def split_examples(examples, test_split=0.10, dev_split=0.10):\n",
+    "    subsets = {}\n",
+    "\n",
+    "    for example in examples:\n",
+    "        dialect = example.expand['dialect'].name.lower()\n",
+    "        if not subsets.get(dialect):\n",
+    "            subsets[dialect] = { 'all': [] }\n",
+    "        subsets[dialect]['all'].append(example)\n",
+    "\n",
+    "    for subset in subsets.values():\n",
+    "        for i, example in enumerate(subset['all']):\n",
+    "            prog = i / len(subset['all'])\n",
+    "\n",
+    "            if prog < test_split:\n",
+    "                split = 'test'\n",
+    "            elif prog < dev_split + test_split:\n",
+    "                split = 'dev'\n",
+    "            else:\n",
+    "                split = 'train'\n",
+    "\n",
+    "            if not subset.get(split):\n",
+    "                subset[split] = []\n",
+    "            subset[split].append(example)\n",
+    "        \n",
+    "        del subset['all']\n",
+    "\n",
+    "    return subsets"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 11,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "subsets = split_examples(examples)"
    ]
   },
   {
@@ -43,39 +102,85 @@
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 24,
    "metadata": {},
    "outputs": [],
    "source": [
     "from pydub import AudioSegment\n",
     "import requests\n",
     "import tempfile\n",
+    "import tarfile\n",
+    "import shutil\n",
+    "import os\n",
+    "import csv\n",
     "\n",
-    "test_split = 0.10\n",
-    "dev_split = 0.10\n",
-    "\n",
-    "for i, example in enumerate(examples):\n",
-    "    prog = i / len(examples)\n",
-    "    \n",
-    "    if prog < test_split:\n",
-    "        split = 'test'\n",
-    "    elif prog < dev_split + test_split:\n",
-    "        split = 'dev'\n",
-    "    else:\n",
-    "        split = 'train'\n",
+    "def save_data(subsets):\n",
+    "    for dialect, subset in subsets.items():\n",
+    "        for split, examples in subset.items():\n",
+    "            audio_dir_path = os.path.join(\"audio\", dialect, split)\n",
+    "            os.makedirs(audio_dir_path, exist_ok=True)\n",
     "\n",
-    "    audio_url = pb.get_file_url(example, example.speech, {})\n",
-    "    response = requests.get(audio_url)\n",
+    "            transcripts = []\n",
+    "            transcript_dir_path = os.path.join(\"transcript\", dialect)\n",
+    "            os.makedirs(transcript_dir_path, exist_ok=True)\n",
+    "            \n",
+    "            for example in examples:\n",
+    "                pb = PocketBase('https://pocketbase.nenadb.dev/')\n",
+    "                audio_url = pb.get_file_url(example, example.speech, {})\n",
+    "                response = requests.get(audio_url)\n",
+    "                with tempfile.NamedTemporaryFile() as f:\n",
+    "                    f.write(response.content)\n",
+    "                    f.flush()\n",
+    "                    audio = AudioSegment.from_file(f.name)\n",
+    "                audio = audio.set_frame_rate(48000)\n",
+    "                audio_file_name = f\"nena_speech_{example.id}.mp3\"\n",
+    "                audio_file_path = os.path.join(audio_dir_path, audio_file_name)\n",
+    "                audio.export(audio_file_path, format=\"mp3\")\n",
+    "                \n",
+    "                transcripts.append({\n",
+    "                    'age': example.age,\n",
+    "                    'transcription': example.transcription,\n",
+    "                    'translation': example.translation,\n",
+    "                    'path': audio_file_name,\n",
+    "                })\n",
     "\n",
-    "    with tempfile.NamedTemporaryFile() as f:\n",
-    "        f.write(response.content)\n",
-    "        f.flush()\n",
-    "        audio = AudioSegment.from_file(f.name)\n",
+    "            audio_tar_path = f\"{audio_dir_path}.tar\"\n",
+    "            with tarfile.open(audio_tar_path, 'w') as tar:\n",
+    "                tar.add(audio_dir_path, arcname=os.path.basename(audio_dir_path))\n",
     "\n",
-    "    audio = audio.set_frame_rate(48000)\n",
-    "    audio.export(f\"nena_speech_{example.id}.mp3\", format=\"mp3\")\n",
+    "            with open(os.path.join(transcript_dir_path, f\"{split}.tsv\"), 'w', newline='') as f:\n",
+    "                writer = csv.DictWriter(f, fieldnames=transcripts[0].keys(), delimiter='\\t')\n",
+    "                writer.writeheader()\n",
+    "                writer.writerows(transcripts)\n",
     "\n",
-    "    break"
+    "            shutil.rmtree(audio_dir_path)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 25,
+   "metadata": {},
+   "outputs": [
+    {
+     "ename": "KeyboardInterrupt",
+     "evalue": "",
+     "output_type": "error",
+     "traceback": [
+      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
+      "\u001b[0;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
+      "\u001b[1;32m/Users/matthew/Documents/nenadb/dataloader/main.ipynb Cell 10\u001b[0m line \u001b[0;36m1\n\u001b[0;32m----> <a href='vscode-notebook-cell:/Users/matthew/Documents/nenadb/dataloader/main.ipynb#X10sZmlsZQ%3D%3D?line=0'>1</a>\u001b[0m save_data(subsets)\n",
+      "\u001b[1;32m/Users/matthew/Documents/nenadb/dataloader/main.ipynb Cell 10\u001b[0m line \u001b[0;36m2\n\u001b[1;32m     <a href='vscode-notebook-cell:/Users/matthew/Documents/nenadb/dataloader/main.ipynb#X10sZmlsZQ%3D%3D?line=23'>24</a>\u001b[0m     f\u001b[39m.\u001b[39mwrite(response\u001b[39m.\u001b[39mcontent)\n\u001b[1;32m     <a href='vscode-notebook-cell:/Users/matthew/Documents/nenadb/dataloader/main.ipynb#X10sZmlsZQ%3D%3D?line=24'>25</a>\u001b[0m     f\u001b[39m.\u001b[39mflush()\n\u001b[0;32m---> <a href='vscode-notebook-cell:/Users/matthew/Documents/nenadb/dataloader/main.ipynb#X10sZmlsZQ%3D%3D?line=25'>26</a>\u001b[0m     audio \u001b[39m=\u001b[39m AudioSegment\u001b[39m.\u001b[39;49mfrom_file(f\u001b[39m.\u001b[39;49mname)\n\u001b[1;32m     <a href='vscode-notebook-cell:/Users/matthew/Documents/nenadb/dataloader/main.ipynb#X10sZmlsZQ%3D%3D?line=26'>27</a>\u001b[0m audio \u001b[39m=\u001b[39m audio\u001b[39m.\u001b[39mset_frame_rate(\u001b[39m48000\u001b[39m)\n\u001b[1;32m     <a href='vscode-notebook-cell:/Users/matthew/Documents/nenadb/dataloader/main.ipynb#X10sZmlsZQ%3D%3D?line=27'>28</a>\u001b[0m audio_file_name \u001b[39m=\u001b[39m \u001b[39mf\u001b[39m\u001b[39m\"\u001b[39m\u001b[39mnena_speech_\u001b[39m\u001b[39m{\u001b[39;00mexample\u001b[39m.\u001b[39mid\u001b[39m}\u001b[39;00m\u001b[39m.mp3\u001b[39m\u001b[39m\"\u001b[39m\n",
+      "File \u001b[0;32m~/Documents/nenadb/dataloader/venv/lib/python3.11/site-packages/pydub/audio_segment.py:728\u001b[0m, in \u001b[0;36mAudioSegment.from_file\u001b[0;34m(cls, file, format, codec, parameters, start_second, duration, **kwargs)\u001b[0m\n\u001b[1;32m    726\u001b[0m     info \u001b[39m=\u001b[39m \u001b[39mNone\u001b[39;00m\n\u001b[1;32m    727\u001b[0m \u001b[39melse\u001b[39;00m:\n\u001b[0;32m--> 728\u001b[0m     info \u001b[39m=\u001b[39m mediainfo_json(orig_file, read_ahead_limit\u001b[39m=\u001b[39;49mread_ahead_limit)\n\u001b[1;32m    729\u001b[0m \u001b[39mif\u001b[39;00m info:\n\u001b[1;32m    730\u001b[0m     audio_streams \u001b[39m=\u001b[39m [x \u001b[39mfor\u001b[39;00m x \u001b[39min\u001b[39;00m info[\u001b[39m'\u001b[39m\u001b[39mstreams\u001b[39m\u001b[39m'\u001b[39m]\n\u001b[1;32m    731\u001b[0m                      \u001b[39mif\u001b[39;00m x[\u001b[39m'\u001b[39m\u001b[39mcodec_type\u001b[39m\u001b[39m'\u001b[39m] \u001b[39m==\u001b[39m \u001b[39m'\u001b[39m\u001b[39maudio\u001b[39m\u001b[39m'\u001b[39m]\n",
+      "File \u001b[0;32m~/Documents/nenadb/dataloader/venv/lib/python3.11/site-packages/pydub/utils.py:275\u001b[0m, in \u001b[0;36mmediainfo_json\u001b[0;34m(filepath, read_ahead_limit)\u001b[0m\n\u001b[1;32m    273\u001b[0m command \u001b[39m=\u001b[39m [prober, \u001b[39m'\u001b[39m\u001b[39m-of\u001b[39m\u001b[39m'\u001b[39m, \u001b[39m'\u001b[39m\u001b[39mjson\u001b[39m\u001b[39m'\u001b[39m] \u001b[39m+\u001b[39m command_args\n\u001b[1;32m    274\u001b[0m res \u001b[39m=\u001b[39m Popen(command, stdin\u001b[39m=\u001b[39mstdin_parameter, stdout\u001b[39m=\u001b[39mPIPE, stderr\u001b[39m=\u001b[39mPIPE)\n\u001b[0;32m--> 275\u001b[0m output, stderr \u001b[39m=\u001b[39m res\u001b[39m.\u001b[39;49mcommunicate(\u001b[39minput\u001b[39;49m\u001b[39m=\u001b[39;49mstdin_data)\n\u001b[1;32m    276\u001b[0m output \u001b[39m=\u001b[39m output\u001b[39m.\u001b[39mdecode(\u001b[39m\"\u001b[39m\u001b[39mutf-8\u001b[39m\u001b[39m\"\u001b[39m, \u001b[39m'\u001b[39m\u001b[39mignore\u001b[39m\u001b[39m'\u001b[39m)\n\u001b[1;32m    277\u001b[0m stderr \u001b[39m=\u001b[39m stderr\u001b[39m.\u001b[39mdecode(\u001b[39m\"\u001b[39m\u001b[39mutf-8\u001b[39m\u001b[39m\"\u001b[39m, \u001b[39m'\u001b[39m\u001b[39mignore\u001b[39m\u001b[39m'\u001b[39m)\n",
+      "File \u001b[0;32m/opt/homebrew/Cellar/python@3.11/3.11.5/Frameworks/Python.framework/Versions/3.11/lib/python3.11/subprocess.py:1209\u001b[0m, in \u001b[0;36mPopen.communicate\u001b[0;34m(self, input, timeout)\u001b[0m\n\u001b[1;32m   1206\u001b[0m     endtime \u001b[39m=\u001b[39m \u001b[39mNone\u001b[39;00m\n\u001b[1;32m   1208\u001b[0m \u001b[39mtry\u001b[39;00m:\n\u001b[0;32m-> 1209\u001b[0m     stdout, stderr \u001b[39m=\u001b[39m \u001b[39mself\u001b[39;49m\u001b[39m.\u001b[39;49m_communicate(\u001b[39minput\u001b[39;49m, endtime, timeout)\n\u001b[1;32m   1210\u001b[0m \u001b[39mexcept\u001b[39;00m \u001b[39mKeyboardInterrupt\u001b[39;00m:\n\u001b[1;32m   1211\u001b[0m     \u001b[39m# https://bugs.python.org/issue25942\u001b[39;00m\n\u001b[1;32m   1212\u001b[0m     \u001b[39m# See the detailed comment in .wait().\u001b[39;00m\n\u001b[1;32m   1213\u001b[0m     \u001b[39mif\u001b[39;00m timeout \u001b[39mis\u001b[39;00m \u001b[39mnot\u001b[39;00m \u001b[39mNone\u001b[39;00m:\n",
+      "File \u001b[0;32m/opt/homebrew/Cellar/python@3.11/3.11.5/Frameworks/Python.framework/Versions/3.11/lib/python3.11/subprocess.py:2108\u001b[0m, in \u001b[0;36mPopen._communicate\u001b[0;34m(self, input, endtime, orig_timeout)\u001b[0m\n\u001b[1;32m   2101\u001b[0m     \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_check_timeout(endtime, orig_timeout,\n\u001b[1;32m   2102\u001b[0m                         stdout, stderr,\n\u001b[1;32m   2103\u001b[0m                         skip_check_and_raise\u001b[39m=\u001b[39m\u001b[39mTrue\u001b[39;00m)\n\u001b[1;32m   2104\u001b[0m     \u001b[39mraise\u001b[39;00m \u001b[39mRuntimeError\u001b[39;00m(  \u001b[39m# Impossible :)\u001b[39;00m\n\u001b[1;32m   2105\u001b[0m         \u001b[39m'\u001b[39m\u001b[39m_check_timeout(..., skip_check_and_raise=True) \u001b[39m\u001b[39m'\u001b[39m\n\u001b[1;32m   2106\u001b[0m         \u001b[39m'\u001b[39m\u001b[39mfailed to raise TimeoutExpired.\u001b[39m\u001b[39m'\u001b[39m)\n\u001b[0;32m-> 2108\u001b[0m ready \u001b[39m=\u001b[39m selector\u001b[39m.\u001b[39;49mselect(timeout)\n\u001b[1;32m   2109\u001b[0m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_check_timeout(endtime, orig_timeout, stdout, stderr)\n\u001b[1;32m   2111\u001b[0m \u001b[39m# XXX Rewrite these to use non-blocking I/O on the file\u001b[39;00m\n\u001b[1;32m   2112\u001b[0m \u001b[39m# objects; they are no longer using C stdio!\u001b[39;00m\n",
+      "File \u001b[0;32m/opt/homebrew/Cellar/python@3.11/3.11.5/Frameworks/Python.framework/Versions/3.11/lib/python3.11/selectors.py:415\u001b[0m, in \u001b[0;36m_PollLikeSelector.select\u001b[0;34m(self, timeout)\u001b[0m\n\u001b[1;32m    413\u001b[0m ready \u001b[39m=\u001b[39m []\n\u001b[1;32m    414\u001b[0m \u001b[39mtry\u001b[39;00m:\n\u001b[0;32m--> 415\u001b[0m     fd_event_list \u001b[39m=\u001b[39m \u001b[39mself\u001b[39m\u001b[39m.\u001b[39m_selector\u001b[39m.\u001b[39mpoll(timeout)\n\u001b[1;32m    416\u001b[0m \u001b[39mexcept\u001b[39;00m \u001b[39mInterruptedError\u001b[39;00m:\n\u001b[1;32m    417\u001b[0m     \u001b[39mreturn\u001b[39;00m ready\n",
+      "\u001b[0;31mKeyboardInterrupt\u001b[0m: "
+     ]
+    }
+   ],
+   "source": [
+    "save_data(subsets)"
    ]
   }
  ],

nena_speech_1_0.py

@@ -0,0 +1,11 @@
+""" NENA Speech Dataset"""
+
+
+import datasets
+
+class NENASpeechConfig(datasets.BuilderConfig):
+    """BuilderConfig for NENASpeech."""
+    pass
+
+class NENASpeech(datasets.GeneratorBasedBuilder):
+    pass

requirements.txt

@@ -1,6 +1,7 @@
-torchaudio
-torch
-pocketbase
 datasets
+pocketbase
 pydub
 requests
+torch
+torchaudio
+tqdm