{ "cells": [ { "cell_type": "code", "execution_count": 1, "id": "d2eb9c7b", "metadata": {}, "outputs": [], "source": [ "import numpy as np\n", "import pandas as pd\n", "\n", "%matplotlib inline\n", "\n", "import os\n", "import json\n", "import requests\n", "import urllib3\n", "\n", "from bs4 import BeautifulSoup, SoupStrainer\n", "import time\n", "\n", "import datetime\n", "import tqdm\n", "import glob" ] }, { "cell_type": "code", "execution_count": 2, "id": "94386983", "metadata": {}, "outputs": [], "source": [ "def make_ordered_unique(l):\n", " seen = set()\n", " unique_list = []\n", " for item in l:\n", " if item not in seen:\n", " seen.add(item)\n", " unique_list.append(item)\n", " return unique_list " ] }, { "cell_type": "code", "execution_count": 3, "id": "9f32b8ee", "metadata": {}, "outputs": [], "source": [ "base_dir = \"/mnt/data-ssd-1/data/nz-parl/\"\n", "manifest_file = os.path.join(base_dir, \"manifest.json\")" ] }, { "cell_type": "code", "execution_count": 4, "id": "7b819820", "metadata": {}, "outputs": [ { "data": { "text/plain": [ "9" ] }, "execution_count": 4, "metadata": {}, "output_type": "execute_result" } ], "source": [ "with open(manifest_file, \"r\") as fi:\n", " manifest = json.load(fi)[:9]\n", "len(manifest)" ] }, { "cell_type": "code", "execution_count": 5, "id": "cc825eed", "metadata": {}, "outputs": [ { "data": { "text/plain": [ "'https://www.parliament.nz/en/pb/hansard-debates/rhr/combined/HansD_20220310_20220310'" ] }, "execution_count": 5, "metadata": {}, "output_type": "execute_result" } ], "source": [ "manifest[3][\"url\"]" ] }, { "cell_type": "code", "execution_count": null, "id": "cfe06869", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": 6, "id": "7a866c23", "metadata": {}, "outputs": [], "source": [ "date_string = \"20220310\"" ] }, { "cell_type": "code", "execution_count": 7, "id": "c630ce95", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "found 35 items in metadata for 20220310\n" ] } ], "source": [ "dd = f\"/mnt/data-ssd-1/data/nz-parl/{date_string}\"\n", "with open(os.path.join(dd, \"results.json\"), \"r\") as fi:\n", " metadata = json.load(fi)\n", "print(f\"found {len(metadata)} items in metadata for {date_string}\")" ] }, { "cell_type": "code", "execution_count": 8, "id": "3f313957", "metadata": {}, "outputs": [ { "data": { "text/plain": [ "{'20220317': 0,\n", " '20220316': 1,\n", " '20220315': 2,\n", " '20220310': 3,\n", " '20220309': 4,\n", " '20220308': 5,\n", " '20220303': 6,\n", " '20220302': 7,\n", " '20220301': 8}" ] }, "execution_count": 8, "metadata": {}, "output_type": "execute_result" } ], "source": [ "# find manifest_element for this date\n", "manifest_url_to_index = {url: i for i, url in enumerate([m[\"url\"] for m in manifest])}\n", "manifest_date_to_index = {url.split(\"_\")[-1]: i for url, i in manifest_url_to_index.items()}\n", "manifest_date_to_index" ] }, { "cell_type": "code", "execution_count": 9, "id": "5d2867fa", "metadata": {}, "outputs": [ { "data": { "text/plain": [ "(35, 35)" ] }, "execution_count": 9, "metadata": {}, "output_type": "execute_result" } ], "source": [ "# make sure the top-level manifest and metadata agree\n", "manifest_item = manifest[manifest_date_to_index[date_string]]\n", "res_vids = set([m[\"url\"] for m in metadata])\n", "manifest_vids = set(manifest_item[\"links\"][\"vimeo\"])\n", "assert manifest_vids == res_vids\n", "len(res_vids), len(manifest_vids)" ] }, { "cell_type": "code", "execution_count": null, "id": "cf206a2b", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": 2, "id": "774eac6c", "metadata": {}, "outputs": [], "source": [ "from suno_utils.utils.conversion import load_audio_as_array, write_wav, convert_audio\n", "from suno_utils.utils.notebook import play_audio, play_array\n", "from IPython.display import Audio\n", "\n", "from suno_utils.utils.conversion import load_audio_as_bytes\n", "from suno_utils.utils.alignment import align_segments, process_align_meta" ] }, { "cell_type": "markdown", "id": "bf255215", "metadata": {}, "source": [ "## Audio files\n", " - ### assumes that mp4s have been converted to mp3s\n", "\n", "### quick and dirty way, in a directory with mp4,\n", " - ### `mkdir audio; for f in *.mp4; ffmpeg -i $f audio/${f:r}.mp3`" ] }, { "cell_type": "code", "execution_count": 11, "id": "f21900e8", "metadata": {}, "outputs": [], "source": [ "# vid_in_order = [\n", "# l.replace(\"https://vimeo.com/\", \"\") + \".mp3\" for l in manifest_item[\"links\"][\"vimeo\"]\n", "# ]\n", "\n", "# files_in_order = make_ordered_unique(vid_in_order)\n", "\n", "# datestr = manifest_item[\"url\"].split(\"_\")[-1]\n", "# files_in_order = [\n", "# os.path.join(base_dir, datestr, \"audio/\", l) for l in files_in_order\n", "# ]\n", "# assert len(files_in_order) == len(manifest_vids)\n", "# print(len(files_in_order))\n", "# files_in_order[:2]\n", "\n", "#manifest[3][\"links\"][\"vimeo\"]" ] }, { "cell_type": "code", "execution_count": 12, "id": "34a0a94d", "metadata": {}, "outputs": [], "source": [ "# concatenate audio file into one mp3\n", "def concat_files(flist):\n", " arr, (old_sr, old_nc) = load_audio_as_array(flist[0])\n", " if old_nc == 2:\n", " arr = arr[:, 0]\n", " pieces = [arr]\n", " for f in tqdm.notebook.tqdm(flist[1:]):\n", " arr, (sample_rate, n_channels) = load_audio_as_array(f)\n", " if sample_rate != old_sr:\n", " raise RuntimeError(\"poo\")\n", " if n_channels == 2:\n", " arr = arr[:, 0]\n", " pieces.append(arr)\n", " return np.concatenate(pieces), sample_rate\n", "\n", "\n", "def write_concatted_mp3(manifest_entry, force=False):\n", " vid_in_order = [\n", " l.replace(\"https://vimeo.com/\", \"\") + \".mp3\"\n", " for l in manifest_entry[\"links\"][\"vimeo\"]\n", " ]\n", "\n", " files_in_order = make_ordered_unique(vid_in_order)\n", " datestr = manifest_entry[\"url\"].split(\"_\")[-1]\n", " audio_dir = os.path.join(\"/mnt/data-ssd-1/data/nz-parl/\", datestr, \"audio/\")\n", " \n", " wav_fname = os.path.join(audio_dir, \"whole-audio.wav\")\n", " mp3_fname = os.path.join(audio_dir, \"whole-audio.mp3\")\n", " \n", " if os.path.exists(mp3_fname) and not force:\n", " print(f\"file {mp3_fname} exists. Skipping...\")\n", " return mp3_fname\n", " \n", " files_in_order = [\n", " os.path.join(audio_dir, l)\n", " for l in files_in_order\n", " ]\n", " files_in_order = list(reversed(files_in_order))\n", " \n", " whole_array, sample_rate = concat_files(files_in_order)\n", " byte_width = whole_array.dtype.itemsize\n", " print(f\"writing wav to {wav_fname}\")\n", " with open(wav_fname, \"wb\") as fi:\n", " write_wav(fi, whole_array.tobytes(), byte_width=byte_width, sample_rate=sample_rate)\n", " \n", " \n", " convert_audio(wav_fname, mp3_fname)\n", " os.remove(wav_fname)\n", " return mp3_fname" ] }, { "cell_type": "code", "execution_count": 14, "id": "25d1e3c1", "metadata": {}, "outputs": [ { "data": { "application/vnd.jupyter.widget-view+json": { "model_id": "d6ea1be15fce4b0e934180d2aab89d9f", "version_major": 2, "version_minor": 0 }, "text/plain": [ " 0%| | 0/9 [00:00\u001b[0;34m\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0mdata_dir\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m", "\u001b[0;31mNameError\u001b[0m: name 'data_dir' is not defined" ] } ], "source": [] }, { "cell_type": "code", "execution_count": null, "id": "58c42a9a", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": 21, "id": "6d5d63df", "metadata": {}, "outputs": [], "source": [ "#transcript[55]" ] }, { "cell_type": "code", "execution_count": 22, "id": "fbf4ffec", "metadata": {}, "outputs": [], "source": [ "# load audio\n", "audio_bytes, (sample_rate, byte_width, _) = load_audio_as_bytes(audio_fname, n_channels=1)\n", "audio_dur_s = len(audio_bytes) // sample_rate // byte_width\n", "if sample_rate != 16_000 or byte_width != 2:\n", " print(\"original file had odd samp rate or byte width\")\n", " audio_bytes_16khz = convert_audio_bytes(\n", " in_bytes=audio_bytes, \n", " in_sample_rate=sample_rate, in_byte_width=byte_width, in_n_channels=1,\n", " out_sample_rate=16_000, out_byte_width=2, out_n_channels=1,\n", " )\n", "else:\n", " audio_bytes_16khz = audio_bytes" ] }, { "cell_type": "code", "execution_count": 23, "id": "f5fbcc18", "metadata": {}, "outputs": [ { "data": { "text/plain": [ "356831" ] }, "execution_count": 23, "metadata": {}, "output_type": "execute_result" } ], "source": [ "full_text = \" \\n \".join(t[\"text\"] for t in transcript)\n", "len(full_text)" ] }, { "cell_type": "code", "execution_count": null, "id": "19dd516b", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": 47, "id": "62484426", "metadata": {}, "outputs": [], "source": [ "import sox\n", "#sox.file_info.duration(audio_fname) / 60" ] }, { "cell_type": "code", "execution_count": null, "id": "0da37aa5", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": null, "id": "39dca9e0", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": 25, "id": "87d13351", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "CPU times: user 482 ms, sys: 984 ms, total: 1.47 s\n", "Wall time: 15min 57s\n" ] } ], "source": [ "%%time\n", "align_dict = align_segments(audio_bytes_16khz, 16_000, 2, 1, full_text)" ] }, { "cell_type": "code", "execution_count": 26, "id": "95918038", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "67.0% aligned\n" ] } ], "source": [ "align_meta = process_align_meta(align_dict[\"words\"])\n", "print(\"{}% aligned\".format(round(np.mean([e[\"success\"] for e in align_meta]) * 100, 1)))" ] }, { "cell_type": "code", "execution_count": 48, "id": "892e495b", "metadata": {}, "outputs": [ { "data": { "text/plain": [ "2" ] }, "execution_count": 48, "metadata": {}, "output_type": "execute_result" } ], "source": [] }, { "cell_type": "code", "execution_count": 27, "id": "e075abd1", "metadata": {}, "outputs": [], "source": [ "#pd.Series([e[\"success\"] for e in align_meta]).astype(float).plot(figsize=(12,8))" ] }, { "cell_type": "code", "execution_count": 28, "id": "1b0e8e64", "metadata": {}, "outputs": [], "source": [ "#pd.Series([e[\"success\"] for e in align_meta]).astype(float).iloc[100:6000].plot(figsize=(12,8))" ] }, { "cell_type": "code", "execution_count": 29, "id": "6ebd4db4", "metadata": {}, "outputs": [ { "data": { "text/plain": [ "0.0" ] }, "execution_count": 29, "metadata": {}, "output_type": "execute_result" } ], "source": [ "np.mean([float(m[\"remote\"]) for m in transcript])" ] }, { "cell_type": "code", "execution_count": 30, "id": "89e5f525", "metadata": {}, "outputs": [ { "data": { "text/plain": [ "{'text': \"I thought irony wasn't allowed!\",\n", " 'strong_text': ['Hon Simon Bridges'],\n", " 'section_header': ['Interjection'],\n", " 'remote': False}" ] }, "execution_count": 30, "metadata": {}, "output_type": "execute_result" } ], "source": [ "transcript[116]" ] }, { "cell_type": "code", "execution_count": null, "id": "3fbebbd8", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": null, "id": "707affa9", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": 31, "id": "c27ebb36", "metadata": {}, "outputs": [], "source": [ "from suno_utils.utils.slicer import (\n", " get_align_gaps,\n", " get_vad_gaps,\n", " get_intervals_from_gaps,\n", " get_audio_slice\n", ")\n", "from suno_utils.utils.alignment import _reconstitute_slice_text\n", "import portion" ] }, { "cell_type": "code", "execution_count": 32, "id": "9dff2499", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "58.1% slice coverage\n" ] } ], "source": [ "# make slices\n", "vad_gap_intervals = get_vad_gaps(audio_bytes_16khz)\n", "align_gap_intervals = get_align_gaps(align_meta, audio_dur_s)\n", "safe_gap_intervals = vad_gap_intervals & align_gap_intervals\n", "slice_intervals = get_intervals_from_gaps(safe_gap_intervals)\n", "cov_frac = np.sum([to_s - from_s for from_s, to_s in slice_intervals]) / audio_dur_s\n", "print(\"{}% slice coverage\".format(round(cov_frac * 100, 1)))\n" ] }, { "cell_type": "code", "execution_count": null, "id": "5b18e58a", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": null, "id": "d736e46e", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": 33, "id": "979e96e8", "metadata": {}, "outputs": [], "source": [ "speaker_idx_map = portion.IntervalDict()\n", "speaker_idx_map[portion.closedopen(0, len(full_text))] = \"speaker\"\n", "meta_idx_map = dict()\n", "\n", "def postprocess_slice_meta2(slice_intervals, align_meta, fulltext):\n", " # annotate with text and speakers\n", " slice_meta = [] # [(start, end), turns]\n", " for start_s, end_s in slice_intervals:\n", " # get character index spans\n", " start_idx = None\n", " for e in align_meta:\n", " if \"start\" in e and e[\"start\"] >= start_s and start_idx is None:\n", " start_idx = e[\"startOffset\"]\n", " end_idx = None\n", " for e in align_meta[::-1]:\n", " if \"end\" in e and e[\"end\"] <= end_s and end_idx is None:\n", " end_idx = e[\"endOffset\"]\n", " # one can be none if edge but not both\n", " if start_idx is None and end_idx is None:\n", " raise ValueError(\"both start and end index are None\")\n", " # assemble speaker segments within index range (if one is none then empty segments)\n", " speaker_turns = []\n", " if start_idx is not None and end_idx is not None:\n", " for (intervals, speaker,) in speaker_idx_map[portion.closedopen(start_idx, end_idx)].items():\n", " for interval in intervals:\n", " text = _reconstitute_slice_text(\n", " interval.lower, interval.upper, fulltext, meta_idx_map\n", " )\n", " speaker_turns.append((speaker, text))\n", " slice_meta.append(((round(start_s, 2), round(end_s, 2)), speaker_turns))\n", " return slice_meta" ] }, { "cell_type": "code", "execution_count": 34, "id": "eef74667", "metadata": {}, "outputs": [], "source": [ "slices_meta = postprocess_slice_meta2(slice_intervals, align_meta, full_text)\n", "slices_audio = [\n", " get_audio_slice(audio_bytes, sample_rate, byte_width, from_s=from_s, to_s=to_s)\n", " for from_s, to_s in slice_intervals\n", "]" ] }, { "cell_type": "code", "execution_count": 35, "id": "dcfbc27b", "metadata": {}, "outputs": [], "source": [ "from suno_utils.utils.notebook import play_bytes" ] }, { "cell_type": "code", "execution_count": null, "id": "b7814372", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": null, "id": "6efecbe7", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": null, "id": "53f54cad", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": 37, "id": "21b23758", "metadata": {}, "outputs": [ { "data": { "text/plain": [ "1308" ] }, "execution_count": 37, "metadata": {}, "output_type": "execute_result" } ], "source": [ "len(slices_audio)" ] }, { "cell_type": "code", "execution_count": 40, "id": "056cd7e5", "metadata": {}, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "We have put significant additional resourcing into our Māori health providers in order to try and reach into some of those communities. It is a more challenging group of people to reach\n" ] }, { "data": { "text/html": [ "\n", " \n", " " ], "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" } ], "source": [ "ii = 213\n", "print(slices_meta[ii][1][0][1])\n", "play_bytes(slices_audio[ii], sample_rate=sample_rate, byte_width=byte_width)" ] }, { "cell_type": "code", "execution_count": 45, "id": "8db9269d", "metadata": {}, "outputs": [ { "data": { "text/plain": [ "" ] }, "execution_count": 45, "metadata": {}, "output_type": "execute_result" }, { "data": { "image/png": "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\n", "text/plain": [ "
" ] }, "metadata": { "needs_background": "light" }, "output_type": "display_data" } ], "source": [ "alens = pd.Series([len(slc) for slc in slices_audio]) / sample_rate / byte_width\n", "alens.hist(bins=30)" ] }, { "cell_type": "code", "execution_count": 186, "id": "1f4b4215", "metadata": {}, "outputs": [ { "data": { "text/plain": [ "1" ] }, "execution_count": 186, "metadata": {}, "output_type": "execute_result" } ], "source": [ "1\n" ] }, { "cell_type": "code", "execution_count": null, "id": "5bb7b5b7", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": 109, "id": "574d25a3", "metadata": {}, "outputs": [ { "data": { "text/plain": [ "dict_keys(['transcript', 'words'])" ] }, "execution_count": 109, "metadata": {}, "output_type": "execute_result" } ], "source": [ "align_dict.keys()" ] }, { "cell_type": "code", "execution_count": 116, "id": "fd72b5ce", "metadata": {}, "outputs": [ { "data": { "text/plain": [ "{'end': 77.77,\n", " 'endOffset': 265,\n", " 'start': 77.34,\n", " 'startOffset': 263,\n", " 'word': 'te',\n", " 'success': False}" ] }, "execution_count": 116, "metadata": {}, "output_type": "execute_result" } ], "source": [ "align_meta[51]" ] }, { "cell_type": "code", "execution_count": null, "id": "af4f557a", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": null, "id": "37884348", "metadata": {}, "outputs": [], "source": [] }, { "cell_type": "code", "execution_count": 74, "id": "75e23e3c", "metadata": {}, "outputs": [], "source": [ "byte_width = whole_array.dtype.itemsize\n", "#(sample_rate, byte_width, n_channels)\n", "with open(\"test.wav\", \"wb\") as fi:\n", " write_wav(fi, whole_array.tobytes(), sample_rate=sample_rate)" ] }, { "cell_type": "code", "execution_count": 64, "id": "032e09ed", "metadata": {}, "outputs": [ { "data": { "application/vnd.jupyter.widget-view+json": { "model_id": "26f9f1c6d9f64090aacf206f1df83e10", "version_major": 2, "version_minor": 0 }, "text/plain": [ " 0%| | 0/3 [00:00