add processing notebook

waxholm-phoneme-fairseq.ipynb

@@ -0,0 +1,317 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": 1,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import soundfile as sf\n",
+    "import wave"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 2,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "def smp_headers(filename: str):\n",
+    "    with open(filename, \"rb\") as f:\n",
+    "        f.seek(0)\n",
+    "        raw_headers = f.read(1024)\n",
+    "        raw_headers = raw_headers.rstrip(b'\\x00')\n",
+    "        asc_headers = raw_headers.decode(\"ascii\")\n",
+    "        asc_headers.rstrip('\\x00')\n",
+    "        tmp = [a for a in asc_headers.split(\"\\r\\n\")]\n",
+    "        back = -1\n",
+    "        while abs(back) > len(tmp) + 1:\n",
+    "            if tmp[back] == '=':\n",
+    "                break\n",
+    "            back -= 1\n",
+    "        tmp = tmp[0:back-1]\n",
+    "        return dict(a.split(\"=\") for a in tmp)\n",
+    "\n",
+    "\n",
+    "def smp_read_sf(filename: str):\n",
+    "    headers = smp_headers(filename)\n",
+    "    if headers[\"msb\"] == \"last\":\n",
+    "        ENDIAN = \"LITTLE\"\n",
+    "    else:\n",
+    "        ENDIAN = \"BIG\"\n",
+    "\n",
+    "    data, sr = sf.read(filename, channels=int(headers[\"nchans\"]),\n",
+    "                       samplerate=16000, endian=ENDIAN, start=512,\n",
+    "                       dtype=\"int16\", format=\"RAW\", subtype=\"PCM_16\")\n",
+    "    return (data, sr)\n",
+    "\n",
+    "\n",
+    "def write_wav(filename, arr):\n",
+    "    with wave.open(filename, \"w\") as f:\n",
+    "        f.setnchannels(1)\n",
+    "        f.setsampwidth(2)\n",
+    "        f.setframerate(16000)\n",
+    "        f.writeframes(arr)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 3,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from pathlib import Path"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 6,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "WAXHOLM = \"/Users/joregan/Playing/waxholm\"\n",
+    "OUTPUT = \"/Users/joregan/Playing/waxholm_fairseq\""
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 7,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "SCENES_PATH = Path(WAXHOLM) / \"scenes_formatted\"\n",
+    "OUTPUT_PATH = Path(OUTPUT)\n",
+    "if not OUTPUT_PATH.is_dir():\n",
+    "    OUTPUT_PATH.mkdir()"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 12,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "TRAIN_FILES = []\n",
+    "with open(Path(WAXHOLM) / \"alloktrainfiles\") as trainf:\n",
+    "    for line in trainf.readlines():\n",
+    "        TRAIN_FILES.append(line.strip())\n",
+    "TEST_FILES = []\n",
+    "with open(Path(WAXHOLM) / \"testfiles\") as testf:\n",
+    "    for line in testf.readlines():\n",
+    "        TEST_FILES.append(line.strip())"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 14,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "1835 327\n"
+     ]
+    }
+   ],
+   "source": [
+    "print(len(TRAIN_FILES), len(TEST_FILES))"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 20,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import re\n",
+    "\n",
+    "def get_labels(mixfile):\n",
+    "    labels = \"\"\n",
+    "    saw_label = False\n",
+    "    with open(mixfile) as infile:\n",
+    "        for line in infile.readlines():\n",
+    "            if not saw_label:\n",
+    "                if line.lower().startswith(\"labels:\"):\n",
+    "                    saw_label = True\n",
+    "                    labels = line[7:].strip()\n",
+    "            else:\n",
+    "                if line.startswith(\"FR\"):\n",
+    "                    break\n",
+    "                else:\n",
+    "                    labels = \" \".join([labels, line.strip()])\n",
+    "        labels = re.sub(\"  +\", \" \", labels)\n",
+    "    return labels"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 21,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "'A:H\\'A: pa p: |h J\\'A:Ggv V\\'ILv pap: sm p:v S\\'E: pa H\\'U:R 2Dd\\'EM Bb\\']:TtE0NG Gg\\']:R 2Tt\\'I STt\"A:VE0#STtR`\\\\M p: \\']: p: \\']M J\\'A: Kk\\'AN F\"O#2S`[TtA Tt\\'I F\"IN#H`AM .'"
+      ]
+     },
+     "execution_count": 21,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "get_labels(\"/Users/joregan/Playing/waxholm/scenes_formatted/fp2043/fp2043.16.03.smp.mix\")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 42,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "def segment_label(label, skip_pause=True):\n",
+    "    phones = []\n",
+    "    i = 0\n",
+    "    while i < len(label):\n",
+    "        start_i = i\n",
+    "        end_i = i\n",
+    "        if label[i:i+2] in [\"NG\", \"E0\", \"kl\", \"sm\"]:\n",
+    "            phones.append(label[i:i+2])\n",
+    "            i += 2\n",
+    "        elif label[i:i+2] == \"p:\":\n",
+    "            if not skip_pause:\n",
+    "                phones.append(\"p:\")\n",
+    "            i += 2\n",
+    "        elif label[i:i+1] == \"#\":\n",
+    "            i += 1\n",
+    "        else:\n",
+    "            if label[i:i+1] in [\"'\", \"`\", \"\\\"\", \"2\", \"~\"]:\n",
+    "                i += 1\n",
+    "                end_i += 1\n",
+    "            if label[i+1:i+2] in [\":\", \"3\", \"4\"]:\n",
+    "                end_i += 1\n",
+    "            phones.append(label[start_i:end_i+1])\n",
+    "            i = end_i + 1\n",
+    "    return phones\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 46,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "assert segment_label(\"Bb\\']:TtE0NG\") == ['B', 'b', \"']:\", 'T', 't', 'E0', 'NG']\n",
+    "assert segment_label(\"STt\\\"A:VE0#STtR`\\\\M\") == ['S', 'T', 't', '\"A:', 'V', 'E0', 'S', 'T', 't', 'R', '`\\\\', 'M']\n",
+    "assert segment_label(\"p:v\") == ['v']"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 56,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "def proc_label(label, stress=False):\n",
+    "    def strip_stress(phone, stress):\n",
+    "        if stress:\n",
+    "            return phone\n",
+    "        if phone[0] in [\"'\", \"`\", \"\\\"\"]:\n",
+    "            return phone[1:]\n",
+    "        else:\n",
+    "            return phone\n",
+    "    words = []\n",
+    "    for word in label.split(\" \"):\n",
+    "        if word in [\"p:pa\", \"pap:\", \"p:pap:\", \"pa\"]:\n",
+    "            words.append(\"pa\")\n",
+    "        elif word == \"p:\" or word == \".\":\n",
+    "            continue\n",
+    "        elif word == \"|h\":\n",
+    "            words.append(\"hes\")\n",
+    "        elif word in [\"sm\", \"ha\", \"kl\"]:\n",
+    "            words.append(word)\n",
+    "        else:\n",
+    "            phones = [strip_stress(p, stress) for p in segment_label(word)]\n",
+    "            words.append(\" \".join(phones))\n",
+    "    return(\" | \".join(words)) + \" |\""
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 57,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "A:H'A: pa p: |h J'A:Ggv V'ILv pap: sm p:v S'E: pa H'U:R 2Dd'EM Bb']:TtE0NG Gg']:R 2Tt'I STt\"A:VE0#STtR`\\M p: ']: p: ']M J'A: Kk'AN F\"O#2S`[TtA Tt'I F\"IN#H`AM .\n",
+      "A: H A: | pa | hes | J A: G g v | V I L v | pa | sm | v | S E: | pa | H U: R | 2D d E M | B b ]: T t E0 NG | G g ]: R | 2T t I | S T t A: V E0 S T t R \\ M | ]: | ] M | J A: | K k A N | F O 2S [ T t A | T t I | F I N H A M |\n"
+     ]
+    }
+   ],
+   "source": [
+    "lbl = get_labels(\"/Users/joregan/Playing/waxholm/scenes_formatted/fp2043/fp2043.16.03.smp.mix\")\n",
+    "plbl = proc_label(lbl)\n",
+    "print(lbl)\n",
+    "print(plbl)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 59,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "with open(OUTPUT_PATH / \"train.tsv\", \"w\") as train_tsv,\\\n",
+    "     open(OUTPUT_PATH / \"train.ltr\", \"w\") as train_ltr,\\\n",
+    "     open(OUTPUT_PATH / \"test.tsv\", \"w\") as test_tsv,\\\n",
+    "     open(OUTPUT_PATH / \"test.ltr\", \"w\") as test_ltr:\n",
+    "    for smpfile in SCENES_PATH.glob(\"fp*/*.smp\"):\n",
+    "        mixfile = f\"{smpfile}.mix\"\n",
+    "        if not Path(mixfile).exists():\n",
+    "            continue\n",
+    "        stem = smpfile.stem\n",
+    "        if f\"{stem}.smp\" in TEST_FILES:\n",
+    "            out_tsv = test_tsv\n",
+    "            out_ltr = test_ltr\n",
+    "        else:\n",
+    "            out_tsv = train_tsv\n",
+    "            out_ltr = train_ltr\n",
+    "\n",
+    "        outwav = str(OUTPUT_PATH / f\"{stem}.wav\")\n",
+    "        arr, sr = smp_read_sf(str(smpfile))\n",
+    "        out_tsv.write(f\"{outwav}\\t{len(arr)}\\n\")\n",
+    "        write_wav(outwav, arr)\n",
+    "        label = get_labels(mixfile)\n",
+    "        ltrline = proc_label(label)\n",
+    "        out_ltr.write(ltrline + \"\\n\")\n",
+    "        \n",
+    "        "
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "hf",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.9.15"
+  },
+  "orig_nbformat": 4
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}