hexgrad2025
Kokoro 82M
End-to-end VAEopenstreambench clip
Punching far above its size; the default local voice when you do not have a GPU.
Plate list
64 engines. 46 you can play.
Approach, license, cloning, streaming, language. If a well-known system has no redistributable clip, it still belongs on the list — with a quiet empty well and a link to the official demo.
Open Home Foundation2023
Piper
End-to-end VAEopenstreambench clip
Edge and offline deployment; hundreds of community voices.
eSpeak NG project1995
eSpeak NG
Concatenative / formantopenstreambench clip
Tiny footprint, many languages, the classic machine accent.
CSTR, University of Edinburgh1996
Festival
Concatenative / formantopenbench clip
The system a generation of papers compared against.
Carnegie Mellon University2001
Flite
Concatenative / formantopenstreambench clip
TTS that fit in feature-phone memory.
ElevenLabs2022
ElevenLabs
Codec LMcommercialclonestreamemotionvendor demo
Voice library, cloning, and inline performance tags like [whispers].
OpenAI2023
OpenAI TTS
AR seq2seqcommercialemotionvendor demo
Steerability through instructions; one-vendor convenience.
OpenAI2024
GPT-4o Audio
Spoken dialoguecommercialstreamemotionvendor demo
End-to-end spoken dialogue, not a vocoder bolted onto a chatbot.
Google Cloud2025
Chirp 3 HD
Zero-shot clonecommercialclonestreamemotionvendor demo
The Cloud TTS line that finally sounds like a conversation.
Google Cloud2022
Google Neural2
AR seq2seqcommercialstreamvendor demo
Workhorse Cloud TTS after WaveNet, before Chirp.
Google Cloud / DeepMind2016
Google WaveNet
AR seq2seqcommercialvendor demo
The moment neural vocoders entered production.
Google Cloud2014
Google Standard
HMM / parametriccommercialstreamvendor demo
Cheap, intelligible, obviously synthetic.
Google Cloud2023
Google Studio
AR seq2seqcommercialemotionvendor demo
Audiobook-style Cloud TTS.
Google DeepMind2026
Gemini 3.1 Flash TTS
Spoken dialoguecommercialemotionno clip
Leaderboard-era Gemini speech with SynthID watermarking.
Amazon Web Services2016
Amazon Polly
AR seq2seqcommercialclonestreamemotionno clip
Being the TTS you can turn on with an IAM role.
Amazon Web Services2025
Amazon Nova Sonic
Spoken dialoguecommercialstreamemotionno clip
AWS’s answer to GPT Realtime and Gemini Live.
Microsoft2019
Azure Neural TTS
AR seq2seqcommercialclonestreamemotionno clip
Enterprise custom neural voice and SSML depth.
Cartesia2024
Cartesia Sonic 3.5
Streamingcommercialclonestreamemotionvendor demo
The lowest published time-to-first-audio among flagship APIs.
PlayAI / PlayHT2021
PlayHT
Codec LMcommercialclonestreamemotionno clip
PlayDialog and ultra-low-latency streaming tiers.
Hume AI2024
Hume Octave 2
Spoken dialoguecommercialclonestreamemotionvendor demo
Emotion as a first-class modelling target, not a slider.
MiniMax2024
MiniMax Speech
Codec LMcommercialclonestreamemotionvendor demo
Price-to-expressiveness on multilingual leaderboards.
Inworld AI2024
Inworld TTS
Streamingcommercialclonestreamemotionno clip
Realtime agent voices at a cost that can stay always-on.
Deepgram2024
Deepgram Aura-2
Streamingcommercialstreamvendor demo
The other half of a Deepgram voice stack.
Fish Audio2024
Fish Audio / Fish Speech
Codec LMresearchclonestreamemotionvendor demo
Top open-weight arena rank with a non-permissive license asterisk.
Sesame2025
Sesame CSM
Spoken dialogueopenclonestreamemotionvendor demo
The ‘this doesn’t sound like TTS’ conversational demo.
Rime Labs2023
Rime
Streamingcommercialclonestreamno clip
Not sounding like an audiobook when it reads an account number.
Murf AI2020
Murf
AR seq2seqcommercialcloneemotionvendor demo
Voiceover tools for people who do not want an SDK.
WellSaid Labs2019
WellSaid
AR seq2seqcommercialcloneemotionvendor demo
Training-video and e-learning narration.
Speechify2016
Speechify SIMBA
Codec LMcommercialclonestreamemotionno clip
Consumer reading, then a competitive API.
Resemble AI2019
Resemble
Zero-shot clonecommercialclonestreamemotionno clip
Cloning as a product, then open-sourcing a strong local model.
Descript2020
Descript Overdub
Zero-shot clonecommercialcloneno clip
Making voice cloning a text cursor.
LOVO2019
LOVO / Genny
AR seq2seqcommercialcloneemotionno clip
Marketplace voices for ads and explainers.
NVIDIA2021
NVIDIA Riva
Streamingcommercialclonestreamno clip
On-prem streaming TTS next to the GPU that ran the LLM.
Li et al.2023
StyleTTS 2
Flow / diffusionopencloneemotionvendor demo
Human-level MOS on a single-speaker corpus, and the recipe Kokoro slimmed down.
Kim, Kong, Son (Kakao)2021
VITS
End-to-end VAEopenstreamvendor demo
Making two-stage TTS look overbuilt.
Kim, Kong, Son, Yoon (Kakao)2020
Glow-TTS
Parallel / non-ARopenvendor demo
The flow model VITS grew out of.
Google2017
Tacotron
AR seq2seqresearchvendor demo
Ending the era of hand-built TTS frontends.
Google2017
Tacotron 2
AR seq2seqresearchvendor demo
The canonical two-stage neural TTS recipe.
Microsoft2019
FastSpeech
Parallel / non-ARresearchvendor demo
Making neural TTS fast enough to feel like a vocoder problem again.
Microsoft2020
FastSpeech 2
Parallel / non-ARresearchemotionvendor demo
Controllable parallel TTS without a Tacotron teacher.
CSTR / Nagoya Institute of Technology2002
Merlin / HTS
HMM / parametricresearchstreamvendor demo
The pre-neural baseline every 2017–2019 paper had to beat.
Microsoft2022
NaturalSpeech
Flow / diffusionresearchcloneemotionvendor demo
Claiming human-level MOS, then scaling to zero-shot.
Shanghai Jiao Tong / others2024
F5-TTS
Flow / diffusionresearchcloneemotionvendor demo
Simple recipe, strong zero-shot, a lot of local fine-tunes.
Mehta et al.2024
Matcha-TTS
Flow / diffusionopenstreamvendor demo
The short paper you read before you train a flow TTS.
Microsoft2023
VALL-E
Codec LMresearchcloneemotionvendor demo
Zero-shot cloning as in-context learning.
Suno2023
Bark
Codec LMopencloneemotionvendor demo
[laughs] as a first-class output, not a failure mode.
Coqui2023
XTTS v2
Zero-shot cloneresearchclonestreamvendor demo
The cloning model everyone actually ran in 2024.
Casanova et al. / Coqui2021
YourTTS
Zero-shot cloneopenclonevendor demo
Multilingual zero-shot cloning before XTTS was a product name.
MyShell2023
OpenVoice
Zero-shot cloneopencloneemotionno clip
Granular style control on top of a cloned timbre.
FunAudioLLM / Alibaba2024
CosyVoice 2
Streamingopenclonestreamemotionvendor demo
Open streaming clone quality that is actually usable live.
FireRedTeam2024
FireRedTTS
Zero-shot cloneopencloneemotionvendor demo
Another open clone on the same fair sentence as IndexTTS’s neighbours.
IndexTeam / Bilibili2024
IndexTTS 2
Zero-shot cloneopencloneemotionvendor demo
Making cloned speech fit a picture cut.
Resemble AI2025
Chatterbox
Zero-shot cloneopenclonestreamemotionvendor demo
Open cloning with a license you can actually ship.
SparkAudio2025
Spark-TTS
Codec LMopencloneemotionvendor demo
Open codec-LM TTS without a 7B backbone.
2noise2024
ChatTTS
Spoken dialogueresearchemotionvendor demo
Prosody that sounds like a person on a call.
Canopy Labs2025
Orpheus
Codec LMopenclonestreamemotionno clip
LLM-backbone TTS with a permissive license.
Nari Labs2025
Dia
Spoken dialogueopenemotionno clip
Two people in one pass, with [laughs] still intact.
Alibaba Qwen2025
Qwen3-TTS
Codec LMopenclonestreamemotionno clip
The open all-rounder a lot of 2026 local agents actually use.
Meta2023
Meta MMS / Voicebox
Flow / diffusionresearchclonebench clip
Coverage as the feature: languages nobody else’s catalog has.
Kyutai2024
Kyutai Moshi
Spoken dialogueopenstreamemotionno clip
Overlapping conversation, not turn-taking TTS.
Kyutai2026
Kyutai Pocket TTS
Streamingopenclonestreambench clip
Open voice cloning that actually fits in a laptop CPU budget.
Microsoft2025
VibeVoice
Spoken dialogueresearchcloneemotionvendor demo
Context length as the product.
Mistral2026
Voxtral TTS
Codec LMresearchclonestreamemotionno clip
A frontier lab’s speech model that is not a TTS startup.
James Betker2022
Tortoise
Flow / diffusionopencloneemotionvendor demo
Quality by burning latency on purpose.
64 engines in view · 46 with playable audio on the bench