explainx / blog
Kokoro-82M runs high-quality TTS on CPU via Kokoro-FastAPI — OpenAI-compatible API, 50 voices, sub-5s synthesis on old hardware. HN patterns, single-word workaround, and setup from Ariya Hidayat's guide.
Jun 16, 2026
Ahmad (@TheAhmadOsman) shared VibeThinker 3B on June 16, 2026 — a model that reportedly matches Opus 4.5 (non-nerfed) at just 3B parameters. Built on Qwen 2.5-Coder with RL-based post-training. If the claims hold, this is the milestone that puts frontier-quality coding AI on a single consumer GPU.
Jun 4, 2026
Miso Labs' Miso-TTS is the fastest and most emotive open-source voice foundation model. With 110ms latency, one-shot cloning, and local deployment, it's built for production voice agents that users actually love.
Jul 8, 2026
Reasoning models loop on hard prompts until context dies. Liquid AI's Antidoom surgically fixes the first repeat token with Final Token Preference Optimization. explainx.ai explains FTPO, results, and why near-greedy wins after training.
A 303-point Hacker News thread (July 8, 2026) resurfaced Kokoro-82M — the lightweight TTS model that makes GPU-poor developers feel less left behind. The anchor article: Local, CPU-Friendly, High-Quality TTS with Kokoro by Ariya Hidayat (March 2026), which walks through Kokoro-FastAPI, OpenAI-compatible endpoints, and real CPU timings.
The pitch is simple: realistic speech without sending text to a cloud API, on hardware you already own — including a 12-year-old Intel i7 or Apple M2 with the GPU reserved for LLM inference.
| Aspect | Details |
|---|---|
| Model | Kokoro-82M — ~82M parameters |
| Hardware | CPU-first — no NVIDIA required |
| Voices | ~50 presets — see VOICES.md |
| Languages | English-strong; Mandarin, Hindi, others supported |
| Fastest setup | ghcr.io/remsky/kokoro-fastapi-cpu container (~5 GB with bundled weights) |
| API | OpenAI speech-compatible at /v1 |
| Web UI | localhost:8880/web |
| Known weakness | Single-word / homograph pronunciation — community workarounds exist |
| HN sentiment | Accessibility, Home Assistant, article-to-podcast, browser games — with honest limits |
From Ariya's guide and the remsky/kokoro-fastapi project:
podman run -p 8880:8880 ghcr.io/remsky/kokoro-fastapi-cpu
# or: docker run -p 8880:8880 ghcr.io/remsky/kokoro-fastapi-cpu
Open http://localhost:8880/web to type text and hear output immediately.
Point any OpenAI speech client at the local base URL:
export TTS_API_BASE_URL=http://127.0.0.1:8880/v1
export TTS_VOICE="am_eric"
Sample scripts: github.com/remotebrowser/speak (speak.js / speak.py) — output saves as MP3; plays back if SoX is installed.
Voice selection: set TTS_VOICE to any ID from the official voice list (e.g. am_eric, bf_emma).
Test paragraph (Jupiter gas-giant blurb), am_eric voice, best of 3 runs:
| CPU | Generation time |
|---|---|
| AMD Ryzen 7 8745HS | 1.5 s |
| Apple M2 Pro | 4.5 s |
| Intel Core i7-4770K (2014) | 4.7 s |
Takeaway from the article: if a 12-year-old desktop CPU is acceptable, Kokoro is viable on commodity hardware — not just Apple Silicon or latest Ryzen.
HN commenters note iGPU acceleration (M2 ANE ports, Kokoro-CoreML, ONNX int8) can beat raw CPU numbers further; tts-bench scores show Kokoro still competitive ~1.5 years post-release.
Patterns from the July 8, 2026 thread (69 comments, 1.2M views):
| Use case | How Kokoro fits |
|---|---|
| Accessibility product | IPA pronunciation guides for homographs; GPU-free deployment |
| Article → podcast RSS | Scrape/clean URLs → Kokoro TTS → Apple Podcasts morning drive |
| Home Assistant + Sonos | Wyoming protocol wrapper; event-triggered announcements |
| Browser game (WASM) | ~85 MB WASM or ~300 MB WebGPU build — "legitimately super good" per dev |
| Chrome read-aloud extension | Sentence-level highlighting while Kokoro reads selected text |
| Intercom / door system | Natural local voice without cloud latency |
| ePub audiobooks | Community epub → audio pipelines; occasional mispronunciation (e.g. "Malay") |
| Hermes agent pipeline | Link-to-podcast RSS automation — swapped Edge-TTS for Kokoro same day |
MCP route: kokoro-tts-mcp for agent tool integration.
In-browser: StreamingKokoroJS — 100% client-side (HN mixed reports on reliability).
The most upvoted practical comment on HN: Kokoro stumbles on isolated words — saying "six" may produce something like "ah-six-ah".
Workaround (accessibility product builder):
"The word is: six"Intonation is flatter, but reliable. Maintainers on Discord blame small parameter count; HN notes ElevenLabs can fail similarly on edge cases.
Homograph trick (chess example): say "Knight to f3" to Wispr/Google when "Knight" alone maps to "night" — same pattern across STT/TTS stacks.
| Model / tool | HN note |
|---|---|
| pocket-tts | "Much better" with ripped voices; ONNX int8 ~5x realtime on CPU per some users |
| Qwen3-TTS | Strong local voice cloning |
| F5-TTS | Close to ElevenLabs quality per one reply |
| Piper | macOS Claude Code notification announcer |
| Speaches | OpenAI-compatible; download weights on demand; bundles Whisper STT |
| Kokoro + RVC | Post-process for last ~2% nuance (podcast use cases) |
| NeuML kokoro-base-onnx | ONNX pipeline + permissive tokenizer |
For STT (opposite direction), HN pairs Parakeet + Senko diarization — not Kokoro's lane. See our Miso One real-time TTS guide for latency-first voice agents.
Ariya's closing line is the product pattern:
When combined with a local LLM, a speech synthesis system like this allows you to enjoy listening to LLM answers instead of reading them!
That mirrors Gemma 4 on-device duck demos (Parakeet STT → Gemma → Kokoro TTS) and the broader CPU-LLM movement — privacy, no metered speech API, offline air-gapped setups.
Pairing tips:
Official & community sources
Setup commands, voice IDs, and benchmark numbers follow Ariya Hidayat's March 2026 article and the July 8, 2026 HN thread. Container tags and voice lists may change — verify on Hugging Face and the Kokoro-FastAPI repo before production deploys.