Summary — 2026-06-14

Seventeenth pass — 2026-06-14 (Chatterbox — open-source ElevenLabs killer + feed gap examples)

Context. Anthony flagged that the daily X task is missing ~10 hype stories per day. He pasted 5 example tweets the task missed, and named the goal explicitly: keep him up to date with AI-accelerated advances AND surface what to implement — specifically "the open source ElevenLabs killer." This pass scores Chatterbox (the confirmed candidate) plus the 5 feed examples.

Source of Chatterbox. Surfaced via firecrawl_search for "open source ElevenLabs alternative 2026." Verified via direct GitHub fetch: github.com/resemble-ai/chatterbox. The OmniVoice ninth pass (5/9, conditional) was the previous candidate — Chatterbox is scored against that baseline.

Item	Source	P	Q	S	Total	Move
Chatterbox (Resemble AI) — MIT open-source TTS family: English, Turbo 350M, Multilingual V3 500M (23+ languages incl. Arabic), voice cloning from 10s ref clip, paralinguistic tags `[laugh]` `[cough]`, `pip install chatterbox-tts`, `example_for_mac.py` ships with the repo, multilingual V3 released 2026-06-10. 25k★, 3.3k forks, 17 contributors, actively maintained. Evaluated vs ElevenLabs Turbo v2.5 on Podonos (public, reproducible).	`firecrawl_search` → `github.com/resemble-ai/chatterbox`, verified 2026-06-14	3	3	2	8/9	ACT-ON — IMPLEMENT NOW. See implementation path below.
Google AI Video Editor [unverified name] — Google's free chat-to-edit video tool: "edit videos by simply chatting with it, no timelines, no complex editing panels." Could be Google Vids or a new Labs product. Tool name not confirmed from tweet text alone.	@Waseem__786Ai on X, 14L, 5 bookmarks, 2026-06-14	1	1	2	4/9	Conditional [unverified]. Q capped at 1 per unverified rule. Re-score trigger: verify product name via firecrawl; if Google Vids or a distinct Labs product with real public access, Q promotes to 2 (total 5/9). Watch — a free Google chat-to-edit video tool is a real CapCut competitive event that matters for Articulate client workflows.
Ponytail plugin — VS Code / Cursor coding-agent plugin enforcing YAGNI (You Ain't Gonna Need It): stops AI agents adding unrequested complexity; "best code is code you never wrote."	@bryanonchain on X, 502L, 655 bookmarks, 2026-06-14	1	2	1	4/9	Conditional — wrong lane for now. Real signal (502L, 655 bookmarks — the highest-engagement feed item this pass). Genuine anti-slop for coding-agent workflows. P=1 because Anthony isn't primarily a coder; S=1 because Articulate is not a coding shop. Re-score trigger: if Articulate takes on any software-delivery client work, or Sebastian ships significant agentic code, Ponytail installs and P→2 (total 5/9).

⚡ IMPLEMENT NOW — Chatterbox

Why this beats OmniVoice (ninth pass, 5/9 conditional):

OmniVoice stalled at 5/9 on two blockers: (1) FSL-1.1-ALv2 licence fences commercial use; (2) Q=1 because output fidelity vs ElevenLabs wasn't established. Chatterbox clears both:

Licence: genuinely MIT. No non-compete. No FSL two-year fence. Can bundle into a paid Articulate deliverable on day one.
Fidelity: publicly evaluated. Chatterbox Turbo vs ElevenLabs Turbo v2.5 on Podonos — same standardised evaluation suite, publicly accessible at podonos.com/resembleai/chatterbox-turbo-vs-elevenlabs-turbo. Not a vibe claim.

Why P=3:

GIG Gulf testimonial reels: Anthony pays ElevenLabs Creator ($22/mo) for narration VO. Chatterbox runs local, zero per-character cost, no monthly cap. The GIG project budget recovers $22/mo indefinitely.
Arabic: ChatterboxMultilingualTTS with language_id="ar" is a first-class path. Roche-Debbie MEA outputs and GIG Gulf Arabic testimonials currently have no cheap local TTS option — this is it.
example_for_mac.py ships in the repo — designed for Mac, tested on Mac, no GPU required (MPS support since May 2025). MacPro is the right host; mini's 8 GB ceiling applies to the full model but Turbo (350M) is lighter.

Why Q=3:

25k★, 3.3k forks, 17 contributors, latest commit 2026-06-10 (3 days ago on verification date), 4 active model variants (Turbo · English · Multilingual V2 · Multilingual V3), Podonos evals public. This is a production-class open-source TTS, not an alpha. Paralinguistic tags ([laugh], [chuckle], [cough]) in Turbo are a genuine quality lever for character-voiced GIG reels.

Why S=2 (not 3):

Same reasoning as OmniVoice ninth pass: TTS/STT is the commodity layer. The money in voice AI is in the live-loop / orchestration tier (Vapi, real-time agents). Chatterbox is async local TTS — real but not the premium product surface. S=2 until a client scopes "self-hosted Arabic voice cloning on our infra" as a discrete line item, at which point S→3 (total 9/9).

Implementation path (unblocked today):

pip install chatterbox-tts python example_for_mac.py

Run on MacPro. Voice clone from a 10s ref clip of the GIG Gulf testimonial subject (or Anthony's own voice). A/B against the ElevenLabs Creator output on one reel clip. If fidelity matches → cancel ElevenLabs Creator ($22/mo saved from month 1).

For Arabic GIG Gulf / Roche-Debbie: python from chatterbox.mtl_tts import ChatterboxMultilingualTTS model = ChatterboxMultilingualTTS.from_pretrained(device="mps", t3_model="v3") wav = model.generate(arabic_text, language_id="ar", audio_prompt_path="ref_clip.wav")

CLASS-MATCH against failures.md:

OmniVoice ninth pass — same category, same use case, scored 5/9 conditional because of FSL fence + unproven fidelity. Chatterbox is the version that clears both. The ninth-pass note "P promotes to 3 once the widget is a proven daily-driver AND a local box hosts the TTS without choking" is now a build target not a blocker — the Mac example script and MPS support were always there.
Vendor-doc spot-check (ninth pass / twelfth pass discipline) — applied: licence read as MIT in the repo LICENSE file. No FSL, no non-compete. The check cleared.
Corey Haines / 10-prompt pack pattern — does NOT apply here. Chatterbox is the install, not an inspiration to build our own. It's production-class OSS that directly replaces a paid subscription.

Feed coverage gap — what the 5 example tweets show

Anthony flagged these 5 tweets as missed by the daily task. Pre-scored here to close the loop:

Handle	Content summary	Engagement	Score	Call
@ericosiu	"In 12 months every company will run a Company Brain" — connect Slack/GitHub/HubSpot into one AI knowledge layer. Opinion prediction, no named tool or repo.	9L, 613 impressions	≤3/9	Skip (opinion post, pre-filter rule: no named tool/product).
@HowToPrompt__	"This tool lets you download any video/audio from hundreds of websites. 100% free, open source, no ads." Tool URL in tweet image, name not extractable from text.	35L, 55 bookmarks	2/9 (unverified → Q capped at 1)	Skip. Tool not identified; likely yt-dlp or cobalt.tools (already commodity).
@Marie_Haynes	"Looks like @grok explained this better than I could 🦾" — SEO/Grok crossover. No named tool beyond Grok (existing, scored elsewhere).	10L	≤3/9	Skip.
@Waseem__786Ai	Google free AI video editor, chat-to-edit, no timelines. [unverified name — scored above]	14L, 5 bookmarks	4/9	Conditional [unverified].
@bryanonchain	Ponytail plugin — YAGNI for AI coding agents [scored above]	502L, 655 bookmarks	4/9	Conditional.

Root-cause confirmation: The Chrome-scroll (60s) + keyword-search (Claude OR MCP OR "AI agent" OR Anthropic) method missed all 5 because: (a) none contain the target keywords; (b) the Chrome scroll window is too short to catch 24h of feed. Fix is the OAuth 1.0a home timeline API (reads 24h of all followed accounts regardless of keywords). Updated SKILL.md saved at ~/ClaudeWork/daily-x-hype-radar-SKILL-updated.md — copy to /Users/anthonybooth/Documents/Claude/Scheduled/daily-x-hype-radar/SKILL.md to activate.

Re-score triggers

(a) Chatterbox A/B on GIG reel matches ElevenLabs Creator fidelity → cancel ElevenLabs ($22/mo); P confirmed at 3, total confirmed 8/9.
(b) Arabic output tested on a Roche-Debbie / GIG Gulf clip → if accent/naturalness is acceptable for client deliverable, S promotes to 3 (total 9/9) — self-hosted Arabic TTS on client infra is a product surface nobody in UAE/GCC is selling.
(c) Google AI Video Editor product name confirmed + publicly accessible → re-score at 5/9 conditional; if it integrates into Google Workspace without install, S may rise further.
(d) Ponytail adopted on a software delivery engagement → P→2, S→2, total 5/9 conditional.