More screenshots
Works with the AI Agents you already use
Google TTS Narration
Full narration workflow for Google Cloud Text-to-Speech: text prep, voice choice, SSML delivery, chunked rendering, multi-voice dialogue, and QC.
$5
Google TTS Narration
Example session with this skill installed
Narrate this release note for our audio feed: "On Mar. 5, 2026, Dr. Nguyen shipped v2.1 — approx. 40% faster sync for U.S. teams. Read the changelog on our docs site; it's live now, and the old build goes read-only after Fri." One narrator, keep it under 30 seconds, MP3.
- Read your context and instructions
- Compiled the google tts narration
Job branch: single request + SSML (short announcement, per the decision table).
Prep pass — five traps caught:
- "Mar. 5, 2026" → "March fifth, twenty twenty-six" (spelled out so it isn't read as digits or "Mar")
- "Dr." → "Doctor"; "Nguyen" flagged and test-rendered — the pinned voice reads it acceptably, no respelling needed
- "v2.1" → "version two point one"; "approx." → "approximately"; "40%" → "forty percent"
- "U.S." → "United States" to stop a letter-by-letter read; "Fri." → "Friday"
- Homograph check: "live" is the adjective → respelled "lyve" in the narration script only; "Read" is imperative and unambiguous in context — left, noted
Narration script (SSML):
<speak>On March fifth, twenty twenty-six, Doctor Nguyen shipped version two point one, <break time="400ms"/> with approximately forty percent faster sync for United States teams. <break time="500ms"/> Read the changelog on our docs site. It's <emphasis level="moderate">lyve</emphasis> now, and the old build goes read-only after Friday.</speak>
Render settings: voice pinned to en-US-Neural2-F, rate 0.92, pitch 0. Single request — no chunking, no stitching.
QC: first-pass listen at 1x. "lyve" landed correctly, no digits spelled out, no letter-by-letter reading. Runtime ~24 seconds, inside the 30-second cap. Delivered as MP3 at 128 kbps. Billed characters: under 350 — a re-render would have cost effectively nothing, which is why the prep pass, not the budget, decides quality.
Connects securely to your tools. The creator never sees your data.
About this skill
Synthetic narration fails in predictable places: the engine reads your table markup aloud, spells out a URL, guesses the wrong reading of "lead," and sounds subtly different between chapter one and chapter nine. This skill is the workflow that prevents all of it, built around one principle — ninety percent of bad narration is bad input, not a bad model.
Your agent starts with text preparation: stripping what should never be spoken, expanding abbreviations and numbers by hand, resolving homographs before the engine can guess, and keeping a narration script separate from your readable source so iterations never corrupt the original. Then it picks a voice with a small decision tree (narrator vs. dialogue vs. budget draft), pins the exact voice name so renders stay consistent, and shapes delivery with the four SSML tags that actually matter — breaks, emphasis, prosody, and say-as.
Long documents get a chunking protocol: split at paragraph boundaries under the API's byte limit, numbered files, a chunk log so a bad render costs one chunk instead of the whole document, and lossless ffmpeg concatenation. A two-host podcast pattern covers multi-voice work, with per-turn rendering and conversational stitching. Every job closes with a QC pass — first and last 60 seconds, chunk joins, and a failure table covering 403s, malformed SSML, throttling, voice drift, and runaway reads — plus a pre-delivery quality gate. Cost guidance included: at roughly twelve cents per 5,000 words, re-rendering a bad chunk is always cheaper than shipping a misread.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse All Claude Skills.
- Speech only: the API does not sing, and it is not a voice actor — three or more distinct characters in one piece starts to blur.
- Request latency is seconds, so this suits file production, not real-time conversational voice.
- Voice names and variants change over time; the workflow pins exact names, but a pinned voice can still be retired by the provider eventually.
- Pricing figures are rules of thumb from the workflow's experience and may drift from the provider's current rates.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 24 days ago