Bench for local text-to-speech (TTS) models. Three lenses, on whatever hardware you put it on (paid cloud APIs have their own separate board, see Hosted TTS APIs):
- Speed — cold + warm TTFA (time to first audio), RTFx (realtime speed; higher = faster than realtime), memory, on CPU / CUDA / Apple Silicon
- Listen — every model on every prompt, default voice + voice cloning, with inline audio players, so you can pick a model by ear
- Scores — objective metrics per model: UTMOS (naturalness), WER (intelligibility), SIM (cloning fidelity), scored over the bench prompts via seed-tts-eval-style ASR + speaker-verification. Sortable, with a Default/Cloning toggle.
An objective quality score (NAQ) was prototyped but isn't part of the bench — the v2 features didn't track subjective ranking closely enough to publish, so it was pulled and is being redesigned separately. The bench measures speed; quality is by-ear via the Listen lens.
5uck1ess.github.io/tts-bench — listen to every model, no install. Three lenses:
- Listen — one consolidated gallery with an inline
<audio>player for every model on every prompt, in default voice and voice cloning (each clone sits next to the reference it's imitating). Browse by prompt (compare all models on one sentence) or by model (audition one model across prompts); only one clip plays at a time. Audio is rig-independent, so each sample is sourced once from the highest-fidelity rig and tagged with where it came from. Quality, prosody, and artifacts are obvious in 5 seconds — benchmark tables can't show that. - Speed — per-rig leaderboards (Ryzen 9 9950X3D + RTX 5090, Apple M4, Ryzen + RTX 3090) with cold/warm TTFA, RTFx, and memory, sortable. Pick the box you actually own.
- Scores — objective metrics per model (UTMOS naturalness, WER intelligibility, SIM cloning fidelity), with an interactive top-15 chart, sortable tables, and a Default/Cloning toggle. Human votes remain the preference ground truth; these are objective backstops.
Full per-rig reports (every model × prompt × device, plus by-prompt samples) are linked from the Archive.
Quality is subjective, so the ground truth is your ears. The companion TTS Voting Arena is a public, blind A/B listening test — two clips, no model names shown, pick the one that sounds better. No login, ~5 seconds a vote.
- Default voice — which model sounds more natural?
- Cloning — which clone better matches the reference voice?
Votes feed a live human-preference Elo leaderboard right there on the arena. This is where the "best sounding" and cloning calls above come from — every vote sharpens the ranking.
Requires uv and Python 3.11. ~10-15 min install. Disk for the full set is large: ~39 GB of per-model venvs in the repo, plus ~125 GB of model weights downloaded to your Hugging Face cache (~/.cache/huggingface, not the repo) — ~165 GB all-in. Individual models are far smaller, so installing a subset costs a fraction of that.
# Windows — everything, or just the models you want
.\install.ps1
.\install.ps1 kokoro,piper,miso
python bench.py# macOS / Linux — everything, or just the models you want
./install.sh
./install.sh kokoro piper miso
python bench.pyPass model names to install only those (names = the venvs/<name> slugs, which match the tables below — lowercase, e.g. kokoro, f5tts, chatterbox, miso). A few share one install: neutts covers NeuTTS Air + Nano, chatterbox both ChatterBox variants, vibevoice the 0.5B/1.5B, moss_tts both MOSS checkpoints, inflect both Inflect v2 sizes, hojo both Hojo-TTS-Light sizes, qwentts the 1.7B Base + both 0.6B checkpoints, fish is Fish Speech 1.5. Add scoring (plus scoring_sim on Linux) for the objective-metrics venv. bench.py only runs models whose venv exists, so a partial install benches cleanly — install more models later by re-running with new names.
Interactive feel-test: python speak.py kokoro. One-shot A/B comparison: python compare.py "your phrase". See docs/architecture.md for the runner protocol and how to add a model.
Fastest:
- CPU (Ryzen 9 9950X3D, Windows): sanoTTS Heart-Nano — 50ms warm TTFA, 104× RTFx
- CUDA (RTX 5090): Vaniq-Edge — 30ms warm TTFA, 298× RTFx
- CPU + MPS (Apple M4, 16 GB): Piper — 208ms warm TTFA, 32× RTFx
Best sounding: No objective ranking right now — the NAQ score is paused pending redesign. Open the Demos site and use the Listen lens.
Best cloning — maintainer's by-ear ranking (October 2026):
-
- Fish Audio Drama 3
-
- Higgs Audio v3
-
- Echo-TTS
This is one listener's ranking, not a vote count. For crowd preference see the live arena Elo board; for UTMOS/WER/SIM see the Scores lens.
→ full per-rig results · → full cloning ranking
| Model | Params | Released | Predefined | Cloning | Multilingual | SR | Expressive | License |
|---|---|---|---|---|---|---|---|---|
| Audio8 TTS 0.6B (compiled) | 601M | Aug 2026 | ✓ | ✓ | ✓ (11) | 44.1k | — | Apache 2.0 |
| Audio8 TTS Preview 0.1B | 170M | Aug 2026 | ✓ | ✓ | ✓ (11) | 44.1k | — | Audio8 Community v1.0 |
| Audio8 TTS Preview 0.6B | 601M | Jul 2026 | ✓ | ✓ | ✓ (11) | 44.1k | — | Apache 2.0 |
| Audio8 TTS Realistic 0.6B | 601M | Oct 2026 | ✓ | ✓ | ✓ (23) | 44.1k | — | Apache 2.0 |
| AuroraTTS-100K | 95.6K | Oct 2026 | ✓ (1) | — | — (en) | 22.05k | 2 knobs | Apache 2.0 |
| Hojo-TTS-Light 40M | 84.5M (40.4M LM + 13.3M fine + 30.8M decoder) | Aug 2026 | ✓ (15) | — | ✓ (zh+en) | 24k | — | Apache 2.0 |
| Inflect-Micro v2 | 9.36M | Jul 2026 | ✓ (1) | — | — (en) | 24k | 2 knobs | Apache 2.0 |
| Inflect-Nano v2 | 3.96M | Jul 2026 | ✓ (1) | — | — (en) | 24k | 2 knobs | Apache 2.0 |
| KittenTTS 2 | 1.7B | Sep 2026 | ✓ | ✓ | ✓ (16) | 24k | tags | Stellon Labs Community |
| KittenTTS 2 (C++ CPU) | 1.7B | Sep 2026 | ✓ | — | ✓ (16) | 24k | tags | Stellon Labs Community |
| KittenTTS Nano 0.1 | <100M | Aug 2025 | ✓ | — | — | 24k | — | Apache 2.0 |
| KittenTTS Nano 0.8 | 15M | Feb 2026 | ✓ | — | — | 24k | — | Apache 2.0 |
| Kokoro | 82M | Dec 2024 | ✓ | — | ✓ | 24k | — | Apache 2.0 |
| LFM2.5-Audio 1.5B | 1.5B | Dec 2025 | ✓ (4) | — | — (en) | 24k | — | LFM Open v1.0 |
| LFM3-Audio 2B | 2.0B | Oct 2026 | ✓ (2) | — | ✓ (18) | 24k | — | LFM Open v1.0 |
| LinnetTTS | 41.4M (31.5M LM + 9.9M decoder) | Nov 2026 | ✓ (2) | — | — (en) | 24k | — | Apache 2.0 code / Linnet Weights License |
| LuxTTS | 123M | Jan 2026 | ✓ | — | — | 22.05k | — | MIT |
| Magpie-TTS | 357M | Dec 2025 | ✓ | — | ✓ (9) | 22.05k | emotion voices* | NVIDIA OML |
| Maya1 | 3B | Oct 2025 | ✓ (voice desc) | — | — | 24k | tags + desc | Apache 2.0 |
| MeloTTS | ~52M | Feb 2024 | ✓ | — | ✓ (4) | 44.1k | — | MIT |
| Ming-3.0-TTS 0.5B | 1.48B | Oct 2026 | ✓ (100+) | ✓ | ✓ (7) | 44.1k | tags + desc | MIT |
| Ming-3.0-TTS tiny | 8.98B (MoE, 1.3B active) | Oct 2026 | ✓ (100+) | ✓ | ✓ (7) | 44.1k | tags + desc | MIT |
| Orpheus TTS | 3B | Mar 2025 | ✓ (8) | — | — (en) | 24k | tags | Apache 2.0 |
| OuteTTS 1.0 1B | ~1B | Apr 2025 | ✓ | ✓ | ✓ (12) | 44.1k | — | CC-BY-NC-SA 4.0 + Llama 3.2 |
| Parler-TTS Mini v1 | 878M | Jun 2024 | ✓ (voice desc) | — | — | 44.1k | desc* | Apache 2.0 |
| Phonon 31m | 41.1M (30.8M LM + Mimi) | Oct 2026 | ✓ (1) | — | — (en) | 24k | knob | Apache 2.0 |
| Phonon 40m | 50.5M (40.2M LM + Mimi) | Oct 2026 | ✓ (1) | — | ✓ (5) | 24k | knob | CC BY-NC 4.0 |
| Piper | ~15M | Jan 2023 | ✓ | — | ✓ | 22.05k | — | GPL-3.0 |
| Qwen3-TTS 0.6B CustomVoice | 0.6B | Jan 2026 | ✓ (9) | — | ✓ (10) | 24k | — | Apache 2.0 |
| sanoTTS Amy | 1.46M | Jul 2026 | ✓ (1) | — | — (en) | 22.05k | knob | GPL-3.0 |
| sanoTTS Heart-Nano | 294K | Sep 2026 | ✓ (1) | — | — (en) | 24k | knob | GPL-3.0 |
| Scylla's Band | ~131M | Nov 2026 | ✓ (14) | — | ✓ (6) | 24k | 3 knobs | Apache 2.0 |
| Soprano 1.1 80M | 80M | Jan 2026 | ✓ | — | — | 32k | — | Apache 2.0 |
| SteerTTS | 40.8M | Oct 2026 | ✓ (2,825) | ✓ | — (en) | 24k | 12 knobs + tags | MIT code / CC-BY-NC 4.0 weights |
| SteerTTS (no refiner) | 23.1M | Oct 2026 | ✓ (2,825) | ✓ | — (en) | 24k | 12 knobs + tags | MIT code / CC-BY-NC 4.0 weights |
| Supertonic 3 | 99M | May 2026 | ✓ | — | ✓ (31) | 24k | tags | MIT + OpenRAIL-M |
| SupraTTS 0.1 Beta | 43.5M | Sep 2026 | ✓ (1) | — | — (en) | 22.05k | 3 knobs | MIT |
| Vaniq-Edge | 8.91M | Aug 2026 | ✓ (1) | — | — (en) | 24k | knob | MIT |
| VibeVoice Realtime 0.5B | 0.5B | Dec 2025 | ✓ | — | — | 24k | — | MIT |
| Voxtral 4B TTS | 4B | Nov 2025 | ✓ (20) | ✓ | ✓ | 24k | — | CC-BY-NC 4.0 |
| Model | Params | Released | Predefined | Cloning | Multilingual | SR | Expressive | License |
|---|---|---|---|---|---|---|---|---|
| AuK | 1.53B | Sep 2026 | — | ✓ | ✓ | 24k | — | MIT code / Qwen Research (NC) encoder |
| AuK-Flash | 1.53B | Sep 2026 | — | ✓ | ✓ | 24k | — | MIT code / Qwen Research (NC) encoder |
| Breeze TTS 2 | 3.47B | Aug 2026 | — | ✓ | ✓ (zh+en) | 24k | tags + desc | BreezeBlue Research (NC) |
| ChatterBox | 1.2B | Apr 2025 | — | ✓ | — | 24k | knob | MIT |
| ChatterBox Turbo | 744M | Dec 2025 | — | ✓ | — | 24k | tags* | MIT |
| Coqui XTTS-v2 | 750M | Oct 2023 | — | ✓ | ✓ (17) | 24k | — | CPML (non-commercial) |
| CosyVoice 3 0.5B | 0.5B | Dec 2025 | — | ✓ | ✓ | 24k | desc | Apache 2.0 |
| Dia 1.6B-0626 | 1.6B | Jun 2025 | — | ✓ | — | 44.1k | tags | Apache 2.0 |
| dots.tts (soar) | 2B | Jun 2026 | — | ✓ | ✓ (24) | 48k | — | Apache 2.0 |
| DramaBox | 3.3B | Apr 2026 | — | ✓ | — (en) | 48k | desc | LTX-2 Community (NC) |
| Echo-TTS | ~2.8B | Dec 2025 | — | ✓ | — | 44.1k | tags | CC-BY-NC-SA 4.0 |
| F5-TTS v1 | 330M | Oct 2024 | — | ✓ | ✓ (zh+en) | 24k | — | CC-BY-NC |
| FireRedTTS3 | 2.12B | Aug 2026 | — | ✓ | ✓ (24) | 24k | — | Apache 2.0 |
| Fish Audio Drama 3 (ComfyUI port) | 35.9B (MoE, 3B active) | Oct 2026 | — | ✓ | ✓ | 44.1k | tags + desc | Research (non-commercial) |
| Fish Speech 1.5 | ~500M | Nov 2024 | — | ✓ | ✓ | 44.1k | — | CC-BY-NC-SA 4.0 |
| Fish Speech S2-Pro | 4B | Mar 2026 | — | ✓ | ✓ (80+) | 44.1k | tags | Research (non-commercial) |
| Higgs Audio v3 TTS | 4B | Jun 2026 | — | ✓ | ✓ (100) | 24k | tags | Research (NC) |
| Hojo-TTS-Light 80M | 167M (89.4M LM + 20.3M codec enc + 45.0M dec + 12.0M spk) | Aug 2026 | — | ✓ | ✓ (zh+en) | 24k | — | Apache 2.0 |
| IndexTTS-2 | 1.5B | Jun 2025 | — | ✓ | ✓ (zh+en) | 22.05k | emo-ref + desc + knob | bilibili Model License |
| IndexTTS-2.5 | 0.97B | Aug 2026 | — | ✓ | ✓ (5) | 22.05k | emo-ref + desc + knob | bilibili Model License |
| LongCat-AudioDiT 1B | 1.42B | Mar 2026 | — | ✓ | ✓ (zh+en) | 24k | — | MIT |
| LongCat-AudioDiT 3.5B | 3.83B | Mar 2026 | — | ✓ | ✓ (zh+en) | 24k | — | MIT |
| Mars5-TTS | 1.2B | Jun 2024 | — | ✓ | — | 24k | — | AGPL-3.0 |
| MetaVoice-1B | 1.2B | Feb 2024 | — | ✓ | — | 48k | — | Apache 2.0 |
| MioTTS 0.1B | 0.1B | Feb 2026 | — | ✓ | ✓ (en+ja) | 44.1k | — | Falcon-LLM |
| MioTTS 0.6B | 0.6B | Feb 2026 | — | ✓ | ✓ (en+ja) | 44.1k | — | Apache 2.0 |
| MiraTTS | 0.5B | Dec 2025 | — | ✓ | — | 48k | knob | MIT |
| Miso TTS 8B | 8.2B | May 2026 | — | ✓ | — (en) | 24k | — | Modified MIT |
| MOSS-TTS v1.0 | 8B (Qwen3) | Feb 2026 | — | ✓ | ✓ (20) | 24k | — | Apache 2.0 |
| MOSS-TTS v1.5 | 8B (Qwen3) | May 2026 | — | ✓ | ✓ (31) | 24k | tags (pause) | Apache 2.0 |
| MOSS-TTS-Nano | 100M | Apr 2026 | — | ✓ | ✓ (zh+en) | 48k | — | Apache 2.0 |
| NeuTTS Air | 748M | Sep 2025 | — | ✓ | — | 24k | — | Apache 2.0 |
| NeuTTS Nano | 229M | Dec 2025 | — | ✓ | ✓ (4) | 24k | — | Apache 2.0 |
| OmniVoice | ~1B | Mar 2026 | — | ✓ | ✓ (600+) | 24k | tags* | Apache 2.0 code / CC-BY-NC weights |
| OpenVoice v2 | ~100M | Apr 2024 | — | ✓ | ✓ | 22.05k | knob | MIT |
| Pocket-TTS v4 | 100M | Jan 2026 | — | ✓ | ✓ (7) | 24k | — | Apache 2.0 |
| Qwen3-TTS 0.6B Base | 0.6B | Jan 2026 | — | ✓ | ✓ (10) | 24k | — | Apache 2.0 |
| Qwen3-TTS 1.7B Base | 1.7B | Jan 2026 | — | ✓ | ✓ (10) | 24k | — | Apache 2.0 |
| Qwen3-TTS 1.7B (CUDA-graph) | 1.7B | Jan 2026 | — | ✓ | ✓ (10) | 24k | — | Apache 2.0 weights / MIT wrapper |
| Sesame CSM-1B | 1B | Mar 2025 | — | ✓ | — | 24k | — | Apache 2.0 |
| Sopro V2 Turbo | 120M | Aug 2026 | — | ✓ | ✓ (4) | 24k | — | Apache 2.0 |
| Sopro V2 Turbo (streaming) | 120M | Aug 2026 | — | ✓ | ✓ (4) | 24k | — | Apache 2.0 |
| Step-Audio-EditX | 3B | Oct 2025 | — | ✓ | — | 24k | tags + desc | Apache 2.0 |
| StyleTTS 2 | ~148M | Jun 2023 | — | ✓ | — | 24k | knob | MIT |
| VibeVoice 1.5B | 1.5B | Aug 2025 | — | ✓ | — | 24k | — | MIT |
| VibeVoice 7B | 7B | Sep 2025 | — | ✓ | — | 24k | — | MIT |
| VoxCPM2 | 2B | Apr 2026 | — | ✓ | ✓ (30) | 48k | desc | Apache 2.0 |
| WavTTS | 0.67B | May 2026 | — | ✓ | ✓ (zh+en) | 16k | — | MIT code / CC-BY-NC 4.0 weights |
| ZipVoice | 123M | Jun 2025 | — | ✓ | ✓ (zh+en) | 24k | — | Apache 2.0 |
| Zonos v0.1 | 1.6B | Feb 2025 | — | ✓ | ✓ | 44.1k | emo-ref + knob | Apache 2.0 |
| Zonos2 | 8B (MoE, ~900M active) | Jun 2026 | — | ✓ | — | 44.1k | knob | Apache 2.0 |
Expressive column — what explicit emotion/delivery control the model offers:
tags= inline cues in the text itself ((laughs),[sigh],<laugh>);desc= natural-language style/emotion instructions;knob= numeric or preset parameter (exaggeration, style enum, pitch/speed);emo-ref= emotion conditioned on a separate reference clip or emotion vector;—= none (for cloning models, expression simply follows the reference clip).*= caveat applies. Exact syntax, sources, and caveats per model: docs/expressive-control.md. Note the bench feeds every model the same plain prompts for fairness, so these features are not exercised in any score.
Full per-model gotchas + license details: docs/known-issues.md. Models considered but excluded: docs/considered.md.
Predefined vs Cloning. Predefined models have fixed/selectable speaker voices baked into the weights — they speak with no reference needed. Cloning (zero-shot) models have no voice of their own: they synthesize whatever voice you hand them as a reference clip at inference. Given no reference, a pure zero-shot model falls back to a bundled sample (this bench uses
chris_hemsworth_15s.wav), so its "default voice" is just a clone of that clip. A few models do both (e.g. Voxtral has 20 presets and cloning).
Qwen3-TTS rows.
qwentts_06b_custom(0.6B CustomVoice) is the only Qwen row with real preset voices (9 timbres) and sits on the Default board; itsinstructstyle control is not supported at 0.6B (theqwen-ttspackage ignores it for 0b6 despite the card), hence—in Expressive. The 0.6B Base checkpoint (qwentts_06b), like both 1.7B Base rows, is reference-only — Base has no model-native preset voice, so its no-reference run clones the housechris_hemsworth_15s.wavand it appears under Cloning, never Default. Its cloning lens is disabled by an intermittent decode runaway (identical input yielded 12.9 s of audio on one call and 655.3 s on the next) — see docs/known-issues.md.
Rig availability: Voxtral is Mac (MLX, preset-voice only) + Linux (vLLM, cloning); Fish S2-Pro / MetaVoice / Step-Audio-EditX / Higgs Audio v3 / dots.tts / Zonos2 / Orpheus / CosyVoice 3 are Linux-only (CUDA) — Higgs v3 is the one server-backed model (it runs via a Docker
sgl-omniHTTP server, not an in-process load), dots.tts and Zonos2 require Linux-only dependencies/compiled CUDA kernels, and Orpheus (vLLM) + CosyVoice 3 (cu121 / torch 2.3.1) have no Blackwell-compatible Windows path; Echo-TTS and DramaBox are Windows + Linux (CUDA-only, no CPU/MPS; DramaBox needs ~18 GB VRAM). Fish Audio Drama 3 is Windows + Linux through the third-party ComfyUI-Drama3 port (CUDA-only; experts offloaded to system RAM, so it needs ~48 GB RAM and ~24 GB VRAM) — its row measures that port, not Fish's own server. The rest run on Windows + Linux CUDA, most on CPU/MPS too. Per-rig speed + samples on the Demos site.
60 of the 92 tracked models can clone a voice from a reference clip. Three reference formats supported (wav only / wav + transcript / HF-gated wav). Drop a reference into reference/, then python bench.py --reference reference/myvoice.wav.
Reference-format docs + the blind-vote cloning ranking (28 cloning models, human-preference A/B, frozen at 397 votes; the live arena board now has 2,094 cloning votes): docs/cloning.md.
5uck1ess.github.io/tts-bench/hosted.html has a separate board for 28 paid cloud TTS APIs. Their speed is never ranked against the local models above: a hosted call includes the network and the vendor's queue, and you can't run it on your own hardware. Quality is a different matter: the same prompts and scorers apply, so the Scores page's Default voice view can show hosted APIs beside the local models (Local / Hosted / Both). Cloning stays separate, because the hosted vendors clone a different reference voice.
Measured the same way for every vendor. Each vendor reads the bench prompts (English, French and Spanish) with its documented default voice.
- Speed: time to first speech (first sample above −40 dBFS), measured from the US East Coast. The board shows warm-connection numbers, sampled over several days. HTTP and WebSocket are compared for vendors that offer both.
- Quality: UTMOS, WER and Health, scored with the same pipeline as the local Scores lens.
- Price: dollars per 1M characters, read from each vendor's pricing page.
- Every clip is published.
Cloning lens. 19 of the 28 can clone over their API or console. Each clones the same consented reference voice, published as "Reference voice". Each vendor gets the longest sample it accepts, so a product that needs more audio isn't penalised. The page shows each vendor's documented minimum and recommended length, and SIM scores the clone against the reference. The other 9 are listed with the reason they can't clone on our route: enterprise-only, sales-gated, dashboard-only, or none.
Vendors: ElevenLabs (v4, v4 Turbo) · Cartesia Sonic 3.6 · Gemini 3.8 Flash TTS · Inworld TTS-2 · Speechify SIMBA 3.5 · Fish Audio Drama 3 · Mistral Voxtral TTS 3 · xAI TTS · Rime Coda · AssemblyAI TTS-1 · Soniox TTS v2 · Giggy TTS · StepFun Step-Audio 3 · Smallest.ai Lightning v3.1 Pro · Deepgram Flux · Microsoft MAI-Voice-2.1 · MiniMax speech-2.8-hd (via OpenRouter) · VUI Labs Luna · OpenAI (gpt-4o-mini-tts, gpt-realtime-2.1) · Murf Falcon 2.1 · HeyGen Voice 1 · Bland Speech v3 · Gradium · Async Flash v1.5 · Palabra TTS · Resemble Ultra.
Collector, adapters and methodology: docs/hosted.md and scripts/hosted/. Running it needs your own API keys and spends real money (a full sweep costs a few dollars). Some vendors' terms restrict benchmarking; a vendor that objects is removed.
| Machine | Used for |
|---|---|
| Windows desktop (Ryzen 9 9950X3D / 128 GB / RTX 5090 32 GB) | Windows CPU + CUDA bench rows |
| Linux workstation (Ryzen 9 5900XT / 64 GB / RTX 3090 24 GB, Ubuntu Server 24.04) | Linux CPU + CUDA; the only rig that runs Fish-Speech S2 natively |
| Mac (Apple M4 / 16 GB / M4 GPU) | Mac CPU + MPS bench rows |
If you reproduce on different hardware, file an issue or PR with your results and we'll add a column.
- Full results tables — per-rig, per-prompt, per-model
- Cloning ranking — reference formats + blind-vote ranking (28 cloning models, human-preference A/B)
- Architecture — bench design, runner protocol, adding a model
- Expressive control — which models take emotion tags / style prompts / knobs, with exact syntax
- Known issues — per-model gotchas + per-license table
- Considered but skipped — models evaluated and excluded
- Tasks & pending work — open issues, planned features
- Methodology — what's measured, why cold + warm, why reproducible
- Hosted APIs — the paid-API board: roster, collector, speed sampling, cloning rules
MIT for the bench code in this repo. Each TTS model has its own license — see docs/known-issues.md for the full per-model table.
If this bench saved you a weekend of writing your own:
