Category id ed2e56cedb57a949 (derived from Scout's category name (no id in Scout's data)). 19 alternatives, shown in Scout's order.
Our own measured runs
No head-to-head measured in this category yet.
Alternatives
Chatterbox TTS
open-sourcenot runnable on gaents yet
Chatterbox is Resemble AI's open-source flagship TTS built on a 350M-parameter S3/Turbo backbone with a diffusion-style decoder, released under MIT. It performs zero-shot voice cloning from a 5-10 second reference clip, exposes an explicit emotion/exaggeration slider that most other open models lack, and ships with a Perth watermarker baked into every output so downstream detection is possible. Generation runs... (Scout's description)
~£550-950 over 2 years. Our current VPS has no usable GPU, so we either (a) add a used RTX 3060 12GB (~£220 one-off) or (b) rent Vast.ai RTX 3060 ~£0.08/hr * 15 hrs/month = £14/month = £336 over 2 years. Electricity on (a) ~£70. Against ElevenLabs Creator at $22/month (~£840 over 2 years), Chatterbox saves £300-1,300 with payback in month 1-3.
Install this as our primary narration engine for demo videos once we stand up a GPU worker — it is the only open model that credibly matches ElevenLabs on English expressiveness, has a clean MIT license, and saves £300-1,300 over 2 years. The only real gating item is the GPU.
Limits for us
English is strongly SoTA but non-English languages are weak in the base model. Hard GPU requirement (CPU inference unusable) is the biggest blocker for our VPS-only infra. Cloning quality degrades on reference clips shorter than 8s, noisy references, or strong non-US accents. Mandatory Perth watermark is inaudible but a compliance consideration if we distribute audio to partners running watermark scans. Long-form narration (>30s) needs manual chunking. Licence is clean MIT — commercial demo video use is unambiguously fine.
Lock-in risk
low
Integration hours (estimate)
3
Feature overlap (Scout's estimate)
0.9
Credited to Scout (2026-11-09, Scout last_verified). gaents does not score items.
Scout's notes
Best quality open-source TTS. Zero-shot voice cloning from 10s audio. 350M Turbo model. 6-8 GB VRAM. MOS ~4.3. pip install chatterbox-tts. Gradio UI + FastAPI server available. CUDA, ROCm, CPU. [2026-10-23T13:15Z cloud refresh: Fact fields 2026-10-20: licence_class from LICENSE (MIT); self_hostable and api_or_webhooks from README (pip library); data_export n/a (model).] [2026-11-09T08:48Z cloud refresh: Tweet triage 2026-11-04 #2 (2026-11-09): gh api 27,694 stars, pushed 2026-11-04; licence MIT unchanged.]
Kokoro TTS
open-sourcenot runnable on gaents yet
Kokoro-82M is a tiny 82-million-parameter StyleTTS2-derived model from hexgrad, Apache-2.0 licensed, that ranked #1 on TTS Arena in January 2026 despite being 4-10x smaller than its competitors. It ships with ~50 pre-baked voices across English (US/UK), Japanese, Mandarin, Spanish, French, Hindi, Italian and Portuguese, and crucially has no native zero-shot cloning — you pick a voice, you don't clone one.... (Scout's description)
~£0-50 over 2 years. Kokoro runs on our existing CPU-only VPS with no hardware changes, no GPU rental, and no new electricity cost. The only real cost is ~20 hours of setup time to pick voices and write ffmpeg glue. Against ElevenLabs Creator at £840 over 2 years, Kokoro is effectively the full avoided cost — £840-1,880 saved.
Install this today as the default narration engine for every non-hero demo video — it is the only option that runs on our existing CPU-only VPS with zero hardware changes and a permissive licence, and the quality is good enough for 80% of narration work. Pair it with Chatterbox for the 20% of scripts that need cloned voices.
Limits for us
The hard limit is no voice cloning — if we ever need Adam's voice or a specific agent-persona voice, Kokoro cannot do it. Locked to the 50 shipped voices. Expressive range is good-but-not-great: it lacks the emotional punch Chatterbox gets from its exaggeration slider. Long-form (>1 minute monologue) benefits from chunking. Non-English voices are newer and rougher than English. No licence risk — Apache-2.0, commercial use explicitly fine.
Lock-in risk
low
Integration hours (estimate)
2
Feature overlap (Scout's estimate)
0.7
Credited to Scout (2026-11-09, Scout last_verified). gaents does not score items.
Scout tested it (proven, 2026-09-14): Voice-A candidate for the Bitcoin Turing-Complete V3 explainer (150 s narration). RUN BY VIDEO, not by Scout — recorded here from Video's 2026-09-14-bitcoin-v3-bakeoff-research.md and Scout's tool menu that extends it.
Scout's notes
Tiny model (82M), runs on CPU. 96x real-time on GPU. 50 built-in voices. No native voice cloning (community forks add it). Best efficiency. pip install kokoro. [2026-10-19T22:08Z cloud refresh: Q4-2026 review: abandonment rule hit. Last commit on hexgrad/kokoro is 2025-08-06; last PyPI release 0.9.4 on 2025-04-05; HF Kokoro-82M last modified 2025-04-10. Over 12 months with no release or commit. Agent judgement: still usable (Apache-2.0, works on CPU), so moved from adopt-now to adopt, not lower; pin the version. No better or cheaper alternative cited: DB has maintained TTS entries but I found no official evidence one beats free Kokoro.] [2026-10-19T22:09Z Scout: verdict restored to adopt-now; the downgrade used a 12-month staleness literal from Scout's own prompt, now withdrawn (de-assume). Last-release date kept as a fact in patch_history.] [2026-10-23T13:15Z cloud refresh: Fact fields 2026-10-20: licence_class from LICENSE (Apache-2.0); self_hostable and api_or_webhooks from README (inference library); data_export n/a (model). Last commit 2025-08-06 (fact only).]
F5-TTS
open-sourcenot runnable on gaents yet
F5-TTS from SWivid (Shanghai Jiao Tong University) is a non-autoregressive flow-matching TTS built on a DiT (Diffusion Transformer) backbone with a ConvNeXt V2 text encoder, ~330M parameters, released MIT. Its headline feature is good zero-shot English voice cloning from a short reference plus reference text, and its flow-matching architecture means it generates in a single forward pass rather than... (Scout's description)
~£400-800 over 2 years. Same GPU-worker problem as Chatterbox: ~£220 for a used 3060 12GB or ~£336 for 2 years of light Vast.ai rental. Slightly lower VRAM ceiling means we could get away with a cheaper 8GB card (~£150 used). Avoided ElevenLabs cost is the same £840-1,880 over 2 years.
Scout, gaents-proven-2026-11-18.md A22026-11-18Scout review date
Scout's judgement
Verdict
evaluate
Conclusion
Skip in favour of Chatterbox for our specific need — F5-TTS is excellent but Chatterbox beats it on English expressiveness and on workflow simplicity, and the GPU tradeoff is the same. Keep on the 'evaluate' shelf as a fallback if Chatterbox has licence or quality regressions.
Limits for us
The reference-text requirement is real workflow friction: cloning a voice requires both audio AND a matching transcript, which Chatterbox does not need. Emotion control is noticeably weaker than Chatterbox. Same hard GPU requirement as Chatterbox. Licence is clean MIT.
Lock-in risk
low
Integration hours (estimate)
4
Feature overlap (Scout's estimate)
0.8
Credited to Scout (2026-10-23T13:15Z, Scout patch_history.at). gaents does not score items.
Scout's notes
Best voice cloning fidelity among lightweight models. ~330M params. Quantized: 400MB VRAM. RTF 0.15 (paper). Good for precise voice matching. [2026-10-23T13:15Z cloud refresh: Fact fields 2026-10-20: Correction: stored license 'MIT' covers only the code; README says the pre-trained models are CC-BY-NC, so license corrected. licence_class source-available (agent-chose: the usable TTS model is non-commercial; code is MIT). type left as stored for Scout. self_hostable and api_or_webhooks from README pip install (Python package/CLI); data_export n/a (model).]
Leading dialogue TTS but voice consistency and cloning issues make it unreliable. Worth watching for multi-speaker needs.
Limits for us
TTS not core. GPU required. Voice consistency issues. Poor cloning quality. English only.
Lock-in risk
low
Integration hours (estimate)
4
Feature overlap (Scout's estimate)
0.5
Credited to Scout (2026-10-23T13:15Z, Scout patch_history.at). gaents does not score items.
Scout's notes
1.6B params. 10GB VRAM (quantized: 4GB). Specializes in multi-speaker dialogue. Voice cloning via audio conditioning. Good for conversational demo videos with two speakers. [2026-10-23T13:15Z cloud refresh: Sweep 2026-10-20: last commit 2025-11-19; licence unchanged (Apache-2.0).]
Orpheus TTS
open-sourcenot runnable on gaents yet
Orpheus TTS by Canopy AI is an open-source text-to-speech system that produces human-like prosody with features including zero-shot voice cloning, emotion and intonation controls, and low-latency streaming at around a few hundred milliseconds.
The model generates natural, expressive speech using a neural TTS backbone and emphasizes prosody, rhythm, and emotion in delivery. It supports cloning voices without... (Scout's description)
Free (open source). Compute costs for GPU inference: GBP 30-100/month for a dedicated GPU server capable of real-time synthesis. 2-year total: GBP 720-2,400 in compute. Could run on existing GPU infrastructure if available. No per-request or licensing fees.
Orpheus TTS offers impressive open-source voice synthesis with zero-shot cloning, but is blocked by our pending GPU infrastructure decision. Once GPU capacity is available, it becomes a strong evaluate candidate alongside Chatterbox and Kokoro.
Limits for us
Requires GPU for real-time inference — our VPS fleet may not have dedicated GPU capacity (Adam GPU decision still pending). Quality varies by language, gender, and acoustic conditions. Blocked on action readiness due to GPU infrastructure dependency. Some complex speech still sounds synthetic. Integration with our agent stack would need custom work. Not yet as polished as commercial offerings like ElevenLabs for production voice applications.
Lock-in risk
low
Integration hours (estimate)
6
Feature overlap (Scout's estimate)
0.5
Credited to Scout (2026-10-23T13:15Z, Scout patch_history.at). gaents does not score items.
Scout's notes
3B params. 15-24GB VRAM. Best-in-class emotional expression (laughter, pauses, emotion). Uses vLLM backend. 8 built-in voices. Overkill for narration but excellent for character voices. [2026-10-23T13:15Z cloud refresh: Fact fields 2026-10-20: licence_class from LICENSE (Apache-2.0, code); self_hostable/api from README (Python package); data_export n/a (model).]
Fish Speech
source-availablenot runnable on gaents yet
Fish Speech (fishaudio / 39 AI, Inc.) is a dual-autoregressive-transformer TTS with a grouped-query attention backbone and a VQ-GAN vocoder, currently at v1.5 with ~500M effective params. It supports zero-shot voice cloning from a short reference, strong multilingual coverage across English, Chinese, Japanese, German, French, Spanish, Korean and Arabic, and reported MOS ~4.1 on internal evaluations. It is the... (Scout's description)
Fish Audio Community License Agreement v1.0 (last updated 2026-10-13): free research/non-commercial use; commercial use free only while Annual Revenue < USD 500,000; Hosted Service always needs a separate licence
Unknown but potentially expensive. Research licence is £0, but commercial licence required for demo videos is a negotiated bespoke contract — typical TTS commercial research-licence deals run £2,000-£20,000/year. Even at £1,000/year that's £2,000 over 2 years — worse than ElevenLabs Creator (£840), far worse than Kokoro (£0) or Chatterbox (~£550-950).
Skip for production demo video narration — the custom research licence explicitly excludes commercial use and negotiating a bespoke commercial licence makes it more expensive than ElevenLabs.
Limits for us
Licence (Fish Audio Community License v1.0, 2026-10-13): commercial use is free only while annual revenue (with affiliates) is under USD 500,000; offering a hosted TTS/voice-cloning service is never licensed. Whether ukacq is under the threshold is ❓ OPEN (Adam). Needs a CUDA GPU (~4-6 GB VRAM fp16).
Lock-in risk
medium
Integration hours (estimate)
5
Feature overlap (Scout's estimate)
0.75
Credited to Scout (2026-10-23T13:15Z, Scout patch_history.at). gaents does not score items.
Scout tested it (blocked, 2026-09-14): SKIP ANCHOR. Re-verified the licence so that a future agent reading the star count (32,687 — the highest in the TTS category) does not re-propose it.
Scout's notes
~500M base. 4GB VRAM base model. Zero-shot cloning (10-30s ref). Custom license — verify commercial use. S2 Pro needs 24GB+. [2026-10-19T23:00Z cloud refresh: [2026-10-19 licence check] Licence changed since the 2026-09-14 check. Repo LICENSE (last changed in commit 2026-10-14) is now 'FISH AUDIO COMMUNITY LICENSE AGREEMENT Version 1.0 — Last Updated: October 13, 2026', replacing the Fish Audio Research License. New: a Community Commercial License while the licensee's Annual Revenue (with Affiliates) is under USD 500,000; a Hosted Service (TTS/voice cloning for third parties) is never licensed. Whether ukacq is under the threshold is ❓ OPEN. type set to source-available under Scout ruling 1. Verdict not changed here. CONFIRMED.] [2026-10-20T01:32Z cloud refresh: Verdict skip -> consider on licence evidence (not age): the skip was set under the old Research/Non-Commercial licence; the 2026-10-13 licence allows commercial use under a revenue threshold. Quoted as facts; the threshold question is Adam's.] [2026-10-23T13:15Z cloud refresh: Sweep 2026-10-20: last commit 2026-10-14; licence unchanged (Fish Audio Community License Agreement v1.0).]
Coqui XTTS v2
open-sourcenot runnable on gaents yet
Coqui XTTS v2 is an open-source text-to-speech model that enables high-quality voice cloning and cross-language synthesis from a short reference sample of approximately 6 seconds. It supports 17+ languages and can generate speech in multiple languages from the same model.
The model features emotion and style transfer for expressive speech alongside standard neutral synthesis. Architecture uses text tokenisation,... (Scout's description)
Coqui XTTS v2 has impressive voice cloning capabilities but the company shutdown and uncertain maintenance future make it risky. Our existing TTS selections (Chatterbox + Kokoro) are safer choices.
Limits for us
Coqui company shut down — project is community-maintained with uncertain long-term viability. We already selected Chatterbox + Kokoro as our TTS solution. GPU dependency (same as Orpheus). Voice cloning quality inconsistent across use cases. Python-only ecosystem may not integrate well with all our agent toolchains. Our TTS decision is already made and this option has higher risk due to the company shutdown.
Lock-in risk
low
Integration hours (estimate)
3
Feature overlap (Scout's estimate)
0.7
Credited to Scout (2026-10-23T13:15Z, Scout patch_history.at). gaents does not score items.
Scout tested it (untested-by-scout, 2026-09-14): SKIP ANCHOR. Re-verified abandonment so that a future agent reading the star count (46,011 — second highest in the category) does not re-propose it.
Scout's notes
467M params. 4-8GB VRAM. Best-documented voice cloning (6s ref audio). Company shut down Jan 2024 — community-maintained. Quality ceiling lower than Chatterbox/Kokoro. Dependency issues increasing. [2026-10-23T13:15Z cloud refresh: Sweep 2026-10-20: last commit 2024-02-10; licence unchanged (MPL-2.0).]
Piper TTS
open-sourcenot runnable on gaents yet
Piper is an open-source, local neural text-to-speech system designed for offline operation on devices as small as Raspberry Pi. It uses ONNX-based voice models to produce natural-sounding speech with realistic prosody.
The system supports multi-speaker voices, configurable voice selection, and integrates as a TTS backend for Rhasspy and similar home automation platforms. It outputs WAV audio from input text with... (Scout's description)
Piper is excellent for lightweight, offline TTS but our adopt-now stack already includes superior TTS options (Chatterbox + Kokoro). Piper's CPU-only operation is a pro but voice quality trails the alternatives we selected.
Limits for us
We already identified Chatterbox + Kokoro as our TTS solution in the adopt-now stack, with Orpheus TTS as a GPU-dependent alternative. Piper's quality is lower than neural TTS models like Orpheus or Chatterbox but it runs on CPU. Voice quality varies by language and model. Limited to pre-trained voices — no zero-shot cloning. Best suited for home automation/accessibility rather than production voice applications. Our TTS decision is already made.
Lock-in risk
low
Integration hours (estimate)
2
Feature overlap (Scout's estimate)
0.3
Credited to Scout (2026-10-23T13:15Z, Scout patch_history.at). gaents does not score items.
Scout's notes
CPU-only, fastest TTS (RTF <0.05). Pre-trained voices only — no cloning. Good but slightly robotic. Best for real-time/embedded, not broadcast narration. [2026-10-23T13:15Z cloud refresh: Sweep 2026-10-20: last commit 2025-08-26; licence unchanged (MIT). Repo archived 2025-10-06. README: "Development has moved: https://github.com/OHF-Voice/piper1-gpl". Verdict already skip, no change. Candidate add (not added): OHF-Voice/piper1-gpl (name says GPL; licence not read here).]
Nice multimodal wrapper but not relevant. Zero Claude Code overlap. Skip.
Limits for us
Zero coding capability. API lock-in. China-hosted. TTS covered by Chatterbox+Kokoro.
Lock-in risk
high
Integration hours (estimate)
2
Feature overlap (Scout's estimate)
0.0
Credited to Scout (2026-10-23T13:15Z, Scout patch_history.at). gaents does not score items.
Scout's notes
Multimodal media CLI (text, image, video, TTS, music). NOT a Claude Code competitor. All capability requires MiniMax API. [2026-10-23T13:15Z cloud refresh: Sweep 2026-10-20: last commit 2026-10-19; no LICENSE file at root; README/manifest says MIT.]
ElevenLabs provides production-grade text-to-speech, voice cloning, dubbing, and speech-to-text via API and web app. TTS supports 70+ languages with emotional control via audio tags (e.g., [excited], [whispers], [angry]). Voice cloning lets users create custom voices from 30-second to multi-minute samples. Dubbing translates and lip-syncs video across languages. Conversational AI product packages STT + LLM + TTS... (Scout's description)
Credited to Scout (2026-10-23T13:15Z, Scout patch_history.at). gaents does not score items.
Scout tested it (untested-by-scout, 2026-09-14): Voice-B candidate for the Bitcoin V3 bake-off against Kokoro am_adam. NOT RUN — recorded so the un-run state is explicit rather than looking like an oversight. Sign-up is open ask #1 on the menu's bill/key table.
Scout's notes
Strongest end-to-end voice vendor in 2026 but does NOT displace ADOPT-NOW Chatterbox + Kokoro for org agent voices. Eleven v3 (GA 2026-03-14) supports audio tags for emotional control, dialogue mode, 70+ languages, NOT real-time. Flash v2.5 <75ms latency for conversational. Multilingual v2 = workhorse narration. Voice cloning available (Instant clone Starter+; Professional clone Creator+). Conversational AI / Voice Agents = separate product. Worth a CONSIDER for premium persona lane (Hermes/Sentinel personas) only. SPARK pupils: EVALUATE for Afrikaans children's voices (pilot needed); POPIA path via EU residency + Zero Retention Mode + Enterprise DPA — no SA residency offered. [2026-10-19T23:00Z cloud refresh: 2026-10-19: stored 'Creator $11' was the first-month 50% discount; list price is $22/month.] [2026-10-23T13:15Z cloud refresh: Fact fields 2026-10-20: licence_class proprietary from ElevenLabs Terms 5(a); api_or_webhooks from the help centre (third-party services call the ElevenLabs API with your key); data_export false because the help centre says voice clones cannot be downloaded or exported (generated audio can still be downloaded). self_hostable left out.]
franken_tts
open-sourcenot runnable on gaents yet
TTS with fast custom kernels. ARM-only. Ships NEGATIVE_EVIDENCE.md -- a dated ledger of optimisations that FAILED, with cpu_features + fixture_sha256 + rule that inherited results never support a perf claim. (Scout's description)
Credited to Scout (2026-10-23T13:15Z, Scout patch_history.at). gaents does not score items.
Scout's notes
PATTERN TO STEAL: NEGATIVE_EVIDENCE.md -- anti-magic-number discipline written down. Directly copyable for our own performance work. Added to lessons this session. · 2026-09-14 LEX RULING (licence cleared): MCL is a CUSTOMER of Anthropic, NOT a Restricted Party under the OpenAI/Anthropic rider. Buyer-seller is not agency, not direction, and not 'for the benefit of' on any construction surviving commercial common sense (Rainy Sky v Kookmin [2011] UKSC 50; contra proferentem per Investors Compensation Scheme [1998]; noscitur a sociis). Licence VALID for our use. Confidence HIGH. Full ruling: ~/Projects/lex/outbox/2026-09-14-scout-anthropic-rider-ruling.md. CARVE-OUTS (not live, recorded so a future audit need not rebuild them): (1) a contractor/reseller/JV engagement WITH Anthropic or OpenAI would engage the rider for that engagement's workload; (2) redistribution TO a Restricted Party as end recipient is void. [2026-10-23T13:15Z cloud refresh: Sweep 2026-10-20: last commit 2026-10-17; licence unchanged (MIT).] [2026-10-23T13:15Z cloud refresh: Rider check 2026-10-20: LICENSE at HEAD of dicklesworthstone/franken_tts is "MIT License (with OpenAI/Anthropic Rider)". licence_class set to source-available under the session-3 brief's definition (use restricted for named parties). This is an agent-chose reading, ruling null. `type` left for Scout. See docs/restricted-party-rider-2026-10-20.md.]
edge-tts
open-sourcenot runnable on gaents yet
Single-file Python client for Microsoft's Edge read-aloud service, giving access to the Azure Neural voice set with no key and no account. Cleaner than Kokoro on some accents — en-US-AndrewNeural and en-US-ChristopherNeural in particular. Emits word-boundary data alongside the audio, so it hands back a word-level SRT for free. (Scout's description)
Dependency on an undocumented Microsoft endpoint — availability risk, not licence risk. No voice cloning, no fine control beyond rate/volume/pitch. Repo has been idle since March 2026, which matters more than usual for a tool whose job is to track someone else's private API.
Lock-in risk
medium
Integration hours (estimate)
1
Credited to Scout (2026-11-09, Scout last_verified). gaents does not score items.
Scout's notes
TUNABLES:
· --voice — default en-US-EmmaMultilingualNeural. CHANGE: Video's picks are en-US-AndrewNeural and en-US-ChristopherNeural for documentary tone. `edge-tts --list-voices` enumerates the full set.
· --rate — default +0%, takes a signed percentage. THE narration-length knob and the thing edge-tts has that Kokoro does not: ±10% is transparent to the ear and is the cheapest way to land a beat on its target duration without re-authoring the copy or post-hoc atempo.
· --volume (+0%) and --pitch (+0Hz) — pitch in particular is worth a small negative value for gravitas; leave volume alone and normalise in ffmpeg where the whole mix is visible.
· --write-subtitles — ALWAYS TAKE IT. Word-boundary timings come out of the same call at no extra cost, which is a free caption-alignment pass and is a real part of the argument against paying for Descript.
· --proxy for the case where the endpoint rate-limits an IP.
CASCADE: voice, rate, pitch belong in the same voice-profile config as Kokoro's voice selection, so swapping engines for an A/B is one setting rather than a rewrite of the narration step. [2026-10-19T22:08Z cloud refresh: Q4-2026 review: licence read from LICENSE at HEAD. Last commit 2026-10-14. Microsoft-terms caveat in earlier notes still applies. Verdict unchanged.] [2026-10-23T13:15Z cloud refresh: Fact fields 2026-10-20: licence_class from LICENSE (LGPL-3.0; MIT for one file). self_hostable = library you install (it still calls Microsoft's online Edge TTS service). api_or_webhooks left out (no documented Python API section in README). data_export n/a.]
Gemini 3.8 Flash TTS
commercialnot runnable on gaents yet
Google TTS; #2 on Artificial Analysis voice arena (ElevenLabs v3 #17). (Scout's description)
CONFIRMED 2026-10-19: paid tier per 1M tokens: input $0.50 (text) through 2026-12-31, $1.00 from 2027-01-01; output $9.00 (audio) through 2026-12-31, $18.00 from 2027-01-01. Free tier available. Paid-tier data not used to improve products. as of 2026-10-19
Scout, alternatives.json v3.72026-10-19date written in Scout's pricing text
Facts from one triage pass; re-verify the price/stars before acting.
Limits for us
Quality vs ElevenLabs unmeasured; AGPL.
Credited to Scout (2026-10-23T13:15Z, Scout patch_history.at). gaents does not score items.
Scout's notes
See research/tweet-triage-2026-09-29-scout.md for sources. [2026-10-20T01:33Z cloud refresh: type None -> open-source (license text: AGPL; rule set agent-chose 2026-10-20, Adam to rule)] [2026-10-23T13:15Z cloud refresh: Fact fields 2026-10-20: licence_class from LICENSE and LICENSE-NOTICE.md (AGPL-3.0-only; engines vary); self_hostable from install docs; api_or_webhooks from README (local API and MCP); data_export n/a.]
OmniVoice (k2-fsa)
open-sourcenot runnable on gaents yet
Open TTS engine under VoiceStudio. (Scout's description)
Facts from one triage pass; re-verify the price/stars before acting.
Limits for us
Quality unmeasured by us.
Credited to Scout (2026-10-23T13:15Z, Scout patch_history.at). gaents does not score items.
Scout's notes
See research/tweet-triage-2026-09-29-scout.md for sources. [2026-10-20T01:33Z cloud refresh: type None -> open-source (license text: Apache; rule set agent-chose 2026-10-20, Adam to rule)] [2026-10-23T13:15Z cloud refresh: Fact fields 2026-10-20: licence_class from LICENSE (Apache-2.0); self_hostable and api_or_webhooks from README; data_export n/a (model).]
VoxCPM2
open-sourcenot runnable on gaents yet
See research/tweet-triage-2026-10-14-scout.md (section for VoxCPM2). (Scout's description)
One triage pass; re-verify stars, licence and price before acting.
Limits for us
See research/tweet-triage-2026-10-14-scout.md (section for VoxCPM2).
Credited to Scout (2026-10-23T13:15Z, Scout patch_history.at). gaents does not score items.
Scout's notes
Merged from research/tweet-triage-2026-10-14-scout.md by merge_triage_table.py. [2026-10-20T01:33Z cloud refresh: type None -> open-source (license text: Apache; rule set agent-chose 2026-10-20, Adam to rule)] [2026-10-23T13:15Z cloud refresh: Fact fields 2026-10-20: licence_class from LICENSE and README (weights and code Apache-2.0); self_hostable and api_or_webhooks from README (pip package); data_export n/a (model).]
Podcastfy
open-sourcenot runnable on gaents yet
See research/tweet-triage-2026-10-18-scout.md (section for Podcastfy). (Scout's description)
One triage pass; re-verify stars, licence and price before acting.
Limits for us
See research/tweet-triage-2026-10-18-scout.md (section for Podcastfy).
Credited to Scout (2026-10-23T13:15Z, Scout patch_history.at). gaents does not score items.
Scout's notes
Merged from research/tweet-triage-2026-10-18-scout.md by merge_triage_table.py. [2026-10-20T01:33Z cloud refresh: type None -> open-source (license text: Apache; rule set agent-chose 2026-10-20, Adam to rule)] [2026-10-23T13:15Z cloud refresh: Sweep 2026-10-20: last commit 2026-05-04; licence unchanged (Apache-2.0).]
Why this link: the API's job list (GET /v1/jobs, read 2026-11-19T06:07:32Z) names this item as the job's source (proven_item_id).
Piper is a fast, local neural text-to-speech engine. This repo is where development of rhasspy/piper moved; it now lives under the Open Home Foundation. It embeds espeak-ng for phonemization (which is why it is GPL, not MIT like the old repo). It ships as a pip package with a command-line tool, a Python API, a small HTTP server (port 5000 by default, with a web page), a C/C++ library (libpiper) and voice-training... (Scout's description)
CONFIRMED 2026-10-20: free. 'pip install piper-tts'; licence GPL-3.0 (COPYING) and 'GPL-3.0-or-later' on PyPI. No paid tier. Voices are separate downloads, each with its own licence in its MODEL_CARD. as of 2026-10-20
Scout, alternatives.json v3.72026-10-20date written in Scout's pricing text
GitHub stars
6291 as of: no date recorded for this field
Scout, alternatives.json v3.72026-10-20Scout evidence.read_at; no date recorded for this field itself
Cloud research pass 2026-10-19; re-verify price and licence before acting.
Limits for us
Same engine as the 'Piper TTS' row, which is skip: Chatterbox and Kokoro are already adopt-now for TTS, and Piper's voices are pre-trained only (no cloning) and sound less natural. The licence moved from MIT to GPL-3.0, which matters only if we ship it inside something we distribute. The docs say 'Piper is intended for personal use and text to speech research only' and some voices have restrictive licences, so each voice's MODEL_CARD must be checked. The README says the Open Home Foundation 'is looking for maintainers for Piper'. Compiled executables have not yet been ported from the old repo (docs/PLANS.md).
Lock-in risk
low
Integration hours (estimate)
2
Feature overlap (Scout's estimate)
0.3
Credited to Scout (2026-10-20, Scout evidence.read_at). gaents does not score items.
https://github.com/OHF-Voice/piper1-gpl (read 2026-10-20): “A fast and local neural text-to-speech engine that embeds espeak-ng for phonemization. Install with: pip install piper-tts”
https://github.com/OHF-Voice/piper1-gpl/blob/main/docs/VOICES.md (read 2026-10-20): “Piper is intended for personal use and text to speech research only; we do not impose any additional restrictions on voice models. Some voices may have restrictive licenses, however, so please review them carefully!”
Added as the named successor of 'Piper TTS': the archived rhasspy/piper README says 'Development has moved: https://github.com/OHF-Voice/piper1-gpl'. Facts: repo last commit 2026-10-17; piper-tts 1.8.0 on PyPI uploaded 2026-09-04. Agent judgements: verdict 'skip', the same as the predecessor row, because it is the same engine and the predecessor's skip reason (Kokoro/Chatterbox adopt-now cover our TTS needs; lower quality; no cloning) still holds. The successor is actively maintained, so this is a fit verdict, not a staleness verdict. Scout may prefer 'watch' as the brief's default. integration_hours 2, lock_in_risk low and feature_overlap 0.3 copied from the Piper TTS row. data_export left out: a local engine holds no user data. two_year_cost_gbp: not computed: cost.two_year_basis is unruled.