Founder of Fish Audio @ Hanabi AI Inc. Toward Personalized Voice Agent.
Santa Clara
- Tweets 85
- Following 121
- Followers 536
- Likes 1,212
After reaching 15M+ users, today we're launching Drama 3, the most emotive and controllable TTS model.
$52M seed round. 15M+ users. But this is what we're most proud of yet.
Introducing Drama 3: the most emotive and controllable TTS model in the world
And yes, it's open-weight.
You can now control how even a single word should sound, change its tone, and shift your voice mid-sentence.
Uncle Roger was impressed too.
Try it here:
s.fish.audio/rqh0l8
Comment “DRAMA” within the next 48 hours for 1 month free.
If you're an enterprise your first month is on us:
s.fish.audio/qz9r0t
2
6
370
Lengyue retweeted
Introducing Drama 3 (preview): the most controllable TTS model ever.
Describe the tone, pacing, and character in simple language, without thinking in audio tags.
Shift your voice mid-sentence, generate a multi-character scene, or fix only a single word.
Comment DRAMA to get an API key 🐟
1,296
151
1,952
165,572
I needed a way to know if our model is actually better. So I built a blind test into our production platform and let 5,098 users decide.
We tested S2 Pro and S1 against ElevenLabs V3, ElevenLabs 2.5 Flash, ElevenLabs Multilingual V2, Inworld TTS 1.5 Max, and MiniMax Speech 2.8 HD.
S2 Pro ranked #1 overall, nearly 1.7x the next closest model.
But what stood out most was the per-language gap. In Chinese, S2 Pro scored 8.11. The next closest competitor scored 2.36.
Most TTS companies build for English first and then extend it to fit everything else. Our multilingual data pipeline is why the gap is widest where others invest the least.
Feel free to check out the our report at fish.audio/blog/blind-tts-pr…
Fish Audio performed a 10-day blind test across 5,098 real user comparisons on production traffic. Our S2 Pro model ranked #1 overall with a 66% win rate - outperforming the next closest competitor by over 60%.
- S2 Pro beat ElevenLabs V3 60% to 40%
- S2 Pro beat Inworld 80% to 20%
- S2 Pro beat MiniMax 95% to 5%
- #1 in every language tested
No one knew they were part of a test. They just picked the voice they liked more.
Check full methodology and results here → fish.audio/blog/blind-tts-pr…
9
19
1,653
🎉 Congrats to @FishAudio on launching Fish Audio S2, a frontier TTS model with fine-grained prosody & emotion control via natural-language inline tags. SGLang Day-0 support is now live!
🏆 Best WER on Seed-TTS Eval; 81.88% win rate on EmergentTTS-Eval
🎙️ Voice cloning with 86.4% prefix-cache hit rate via RadixAttention
⚡️ RTF 0.34, 63.3 tok/s on single H200 (single batch)
🌍 Trained on 10M+ hours of audio across ~100 languages, GRPO-aligned
🔧 Dual-AR (Slow + Fast AR) is LLM-isomorphic: continuous batching, paged KV cache & CUDA graphs inherited natively
🗣️ Native multi-speaker: turn-taking, interruptions & cross-speaker emotion in a single pass
👉Cookbook: cookbook.sglang.io/omni/S2-P…
👉Blog: fish.audio/blog/fish-audio-o…
🎬 Curious how to run with SGLang? Check out this voice cloning demo from @GenAI_is_real with Fishaudio-S2-Pro:
Today we launch Fish Audio S2, a new generation of expressive TTS with absurdly controllable emotion.
- open-source
- sub 150ms latency
- multi-speaker in one pass
Real freedom of speech starts now 👇
Community note
The "open source" claim is false. Their custom license prohibits commercial use and limits some derivative uses. A more accurate description is "source-available", meaning the source code is publicly accessible but subject to use restrictions.
opensource.org/osd
github.com/fishaudio/fish…
fossa.com/blog/comprehen
4
12
98
41,922
Incredibly proud of what our team has built — thank you to everyone at Fish Audio for making this happen :D
Today we launch Fish Audio S2, a new generation of expressive TTS with absurdly controllable emotion.
- open-source
- sub 150ms latency
- multi-speaker in one pass
Real freedom of speech starts now 👇
Community note
The "open source" claim is false. Their custom license prohibits commercial use and limits some derivative uses. A more accurate description is "source-available", meaning the source code is publicly accessible but subject to use restrictions.
opensource.org/osd
github.com/fishaudio/fish…
fossa.com/blog/comprehen
4
1
21
1,477
Lengyue retweeted
Niji V7 is now live. This model features improved anime coherence, prompt understanding, text rendering, and sref performance. Enjoy!
386
769
6,736
2,212,916
Lengyue retweeted
Today we’re excited to publicly launch Fish Audio S1, the most expressive and natural TTS model on the market.
- 6x cheaper than elevenlabs
- 20K active developers
- 5M ARR (real ARR)
Clone your own voice for free and get started 👇
249
309
3,570
527,368
Lengyue retweeted
Introducing OpenAudio S1! 🎉
Command your AI voice actor like never before:
- 🔥 Experience unparalleled expressiveness & naturalness: Voted #1 in TTS-Arena!
- 🎯 Achieve state-of-the-art accuracy: 0.008 WER & 0.004 CER in Seed TTS Eval.
- 💬 Control a spectrum of emotions via natural language: From (angry), (happy), (sad) to nuanced (emphasize), (whispering), (empathetic), and more!
- 🎬 Unleash your creative vision: Ideal for Video, Audiobooks, Podcasts, AI Companions, Gaming & endless possibilities!
🔗 Explore OpenAudio S1: fish.audio/
📖 Get the full story: openaudio.com/blogs/s1
113
146
1,263
1,230,776
Lengyue retweeted
We’re excited to release ACE-Step / ACE-Step-v1-3.5B, a fast, versatile DiT-based foundation model for music generation that runs on consumer-grade GPUs.
With its simple architecture and low hardware requirements, it’s easy to fine-tune for various music tasks, empowering, not replacing, artists and creators.
Think of it as a step toward music’s Stable Diffusion moment.
※ Trained on authorized, purchased data.
Demo Page: ace-step.github.io/
Hugging Face: huggingface.co/ACE-Step/ACE-…
Git repo: github.com/ace-step/ACE-Step
32
55
182
113,115
Lengyue retweeted
OpenAI just released AGI (all ghibli images)
Today we launch W25.
Our most powerful batch yet.
They broke the benchmarks for acceleration so much that 0-$3M will never hit the same…and if you want to know why we’re building a ring of servers around the sun, watch carefully.
41
64
387
80,622
Lengyue retweeted
If you haven’t adopted this new way of building companies, you will get outcompeted.
In twelve weeks
- four of ten teams broke $2M ARR
- fifth team hit $1.7M
- top team went from 1 to 10M
Launching applications today. Ten teams. Only one HF0.
54
109
749
163,748
We're deeply grateful to our team for their outstanding work in making this state-of-the-art open-source TTS possible. We're excited to see what innovations the open-source community will create with Fish Speech 1.5.
We're growing! If you're a talented sales professional or engineer interested in shaping the future of voice technology, reach out to us at hr@fish.audio.
3
2
8
1,789
Lengyue retweeted
TEN-agent has @FishAudio extension now. Thanks @LengyueMatrix for the PR.
github.com/TEN-framework/TEN…
2
3
546
Lengyue retweeted
🚨 New powerful open Text to Speech model: Fish Speech 1.4 - trained on 700K hours of speech, multilingual (8 languages)🔥
> Instant Voice Cloning
> Ultra low latency
> ~1GB model weights
> Model weights on the Hub 🤗
> Play with the space in the comments
Kudos to @FishAudio team! They also release a pretty cheap API too 🐐
31
232
1,664
147,323
Lengyue retweeted
Fish Speech：完美支持中英日语言的开源TTS模型
语音处理接近人类水平
模型使用约十五万小时三语数据训练，对中文支持非常的完美。
16
189
874
129,839
Lengyue retweeted
Fish Speech 这个是效果真的不错，生成速度很快，克隆出来的声音也很逼真
测试地址：
fish.audio/zh-CN/text-to-spe…
Fish Audio v1.2 just got released on @huggingface
Very impressive TTS results on Chinese!
huggingface.co/fishaudio/fis…
18
189
732
304,492