Article to speech

/** * Article to speech: turn a web article into an MP3 you can listen to. The article's own paragraphs are taken from the * page (menus, captions and footers left out) and checked word for word against it, then spoken offline in a * session's shell (espeak-ng, installed by the session's setup) and encoded with ffmpeg. No speech API key needed. * *   npx tsx node/index.ts            (BOXLINE_API_KEY; ARTICLE_URL, MAX_WORDS, VOICE, SPEED; a plan with shell sessions) * * Writes output/article.mp3, output/article.txt and output/result.json. */import { mkdirSync, writeFileSync } from "node:fs";import { join } from "node:path";import { Boxline } from "@boxline/sdk";const out = process.env.OUTPUT_DIR ?? "output";const url = process.env.ARTICLE_URL ?? "https://en.wikipedia.org/wiki/Urban_forest";const maxWords = Number(process.env.MAX_WORDS ?? 250);const voice = process.env.VOICE ?? "en-us";const speed = Number(process.env.SPEED ?? 165); // words per minuteconst bx = new Boxline();const norm = (s: string) => s.toLowerCase().replace(/\[\d+\]/g, "").replace(/[^a-z0-9]+/g, " ").trim();// 1. The article's own paragraphs, word for word, and the page's text to check them against.const [r, page] = await Promise.all([  bx.extract<{ title: string; paragraphs: string[] }>({    url,    prompt: `The article on this page: its title, and its paragraphs copied word for word, in order, without menus, captions, footnotes, footers or ads. Stop after about ${maxWords} words, at the end of a paragraph.`,    schema: { type: "object", properties: { title: { type: "string" }, paragraphs: { type: "array", items: { type: "string" } } }, required: ["title", "paragraphs"] },  }),  bx.fetch(url, { format: "text" }),]);const pageText = norm(page.content);// A paragraph that is not on the page as written (the model rewrote it) is not read out.const paragraphs = r.data.paragraphs.filter((p) => pageText.includes(norm(p).slice(0, 120)));const text = [r.data.title, ...paragraphs].join("\n\n");const words = text.split(/\s+/).filter(Boolean).length;console.log(`"${r.data.title}": ${paragraphs.length} of ${r.data.paragraphs.length} paragraphs match the page word for word; ${words} words`);if (!paragraphs.length) throw new Error("no paragraph could be checked against the page");// 2. Speech, made in the machine.const session = await bx.sessions.create({ browser: false, shell: true, setup: ["sudo DEBIAN_FRONTEND=noninteractive apt-get update -qq && sudo DEBIAN_FRONTEND=noninteractive apt-get install -y -qq --no-install-recommends espeak-ng >/dev/null"], timeout: 600, idleTimeout: 300, userMetadata: { example: "article-to-speech" } });console.log(`Session: ${session.id}`);let durationSeconds = 0;try {  await session.files.write("article.txt", text);  const sh = await session.exec(    'espeak-ng -v "$VOICE" -s "$SPEED" -f article.txt -w article.wav && ffmpeg -loglevel error -y -i article.wav -metadata title="$TITLE" -codec:a libmp3lame -qscale:a 5 article.mp3 && ffprobe -v error -show_entries format=duration -of csv=p=0 article.mp3',    { env: { VOICE: voice, SPEED: String(speed), TITLE: r.data.title }, timeoutMs: 300_000 },  );  if (sh.exitCode !== 0) throw new Error(`speech failed: ${sh.stderr.trim()}`);  durationSeconds = Math.round(Number(sh.stdout.trim()) * 10) / 10;  mkdirSync(out, { recursive: true });  writeFileSync(join(out, "article.mp3"), await session.files.read("article.mp3"));  writeFileSync(join(out, "article.txt"), text);} finally {  await session.stop();}console.log(`article.mp3: ${durationSeconds} s (${voice}, ${speed} words per minute)`);writeFileSync(join(out, "result.json"), JSON.stringify({ url, title: r.data.title, paragraphs, dropped: r.data.paragraphs.length - paragraphs.length, words, voice, speed, durationSeconds, usage: { modelUsd: r.usage.costUsd } }, null, 2));

"""Article to speech: turn a web article into an MP3 you can listen to. The article's own paragraphs are taken from thepage (menus, captions and footers left out) and checked word for word against it, then spoken offline in asession's shell (espeak-ng, installed by the session's setup) and encoded with ffmpeg. No speech API key needed.    python python/main.py            (BOXLINE_API_KEY; ARTICLE_URL, MAX_WORDS, VOICE, SPEED; a plan with shell sessions)Writes output/article.mp3, output/article.txt and output/result.json."""import jsonimport osimport refrom concurrent.futures import ThreadPoolExecutorfrom pathlib import Pathfrom boxline import Boxlineout = Path(os.environ.get("OUTPUT_DIR", "output"))url = os.environ.get("ARTICLE_URL", "https://en.wikipedia.org/wiki/Urban_forest")max_words = int(os.environ.get("MAX_WORDS", "250"))voice = os.environ.get("VOICE", "en-us")speed = int(os.environ.get("SPEED", "165"))  # words per minutebx = Boxline()def norm(s: str) -> str:    return re.sub(r"[^a-z0-9]+", " ", re.sub(r"\[\d+\]", "", s.lower())).strip()# 1. The article's own paragraphs, word for word, and the page's text to check them against.with ThreadPoolExecutor(2) as pool:    got = pool.submit(        bx.extract,        url=url,        prompt=f"The article on this page: its title, and its paragraphs copied word for word, in order, without menus, captions, footnotes, footers or ads. Stop after about {max_words} words, at the end of a paragraph.",        schema={"type": "object", "properties": {"title": {"type": "string"}, "paragraphs": {"type": "array", "items": {"type": "string"}}}, "required": ["title", "paragraphs"]},    )    page = pool.submit(bx.fetch, url, format="text").result()    r = got.result()page_text = norm(page["content"])# A paragraph that is not on the page as written (the model rewrote it) is not read out.paragraphs = [p for p in r["data"]["paragraphs"] if norm(p)[:120] in page_text]text = "\n\n".join([r["data"]["title"], *paragraphs])words = len(text.split())print(f"\"{r['data']['title']}\": {len(paragraphs)} of {len(r['data']['paragraphs'])} paragraphs match the page word for word; {words} words")if not paragraphs:    raise SystemExit("no paragraph could be checked against the page")# 2. Speech, made in the machine.setup = ["sudo DEBIAN_FRONTEND=noninteractive apt-get update -qq && sudo DEBIAN_FRONTEND=noninteractive apt-get install -y -qq --no-install-recommends espeak-ng >/dev/null"]with bx.sessions.create(browser=False, shell=True, setup=setup, timeout=600, idle_timeout=300, user_metadata={"example": "article-to-speech"}) as session:    print(f"Session: {session.id}", flush=True)    session.files.write("article.txt", text)    sh = session.exec(        'espeak-ng -v "$VOICE" -s "$SPEED" -f article.txt -w article.wav && ffmpeg -loglevel error -y -i article.wav -metadata title="$TITLE" -codec:a libmp3lame -qscale:a 5 article.mp3 && ffprobe -v error -show_entries format=duration -of csv=p=0 article.mp3',        env={"VOICE": voice, "SPEED": str(speed), "TITLE": r["data"]["title"]},        timeout_ms=300_000,    )    if sh["exitCode"] != 0:        raise SystemExit(f"speech failed: {sh['stderr'].strip()}")    duration = round(float(sh["stdout"].strip()), 1)    out.mkdir(parents=True, exist_ok=True)    (out / "article.mp3").write_bytes(session.files.read("article.mp3"))    (out / "article.txt").write_text(text)print(f"article.mp3: {duration} s ({voice}, {speed} words per minute)")result = {"url": url, "title": r["data"]["title"], "paragraphs": paragraphs, "dropped": len(r["data"]["paragraphs"]) - len(paragraphs), "words": words, "voice": voice, "speed": speed, "durationSeconds": duration, "usage": {"modelUsd": r["usage"]["costUsd"]}}(out / "result.json").write_text(json.dumps(result, indent=2, ensure_ascii=False))

# Node 18 or newer, in the example's folderexport BOXLINE_API_KEY=…   # your API keynpm install @boxline/sdk tsxnpx tsx node/index.ts

# Python 3.9 or newer, in the example's folderexport BOXLINE_API_KEY=…   # your API keypip install boxline-sdkpython python/main.py