Independent evals, research and analysis for voice AI. We evaluate models in the conditions they get deployed in, and work with labs to fix what breaks.
San Francisco
Joined May 2026
- Tweets264
- Following1
- Followers1.4K
- Likes107
Inworld's Realtime TTS 2 ranks #1 of 7 in Customer Support on the Voice Arena Indian English 🇮🇳 TTS Leaderboard, at 1090 Elo.
Realtime TTS 2 from @inworld is built for real-time conversation. Every ranking on Voice Arena comes from blind, head-to-head votes by vetted native speakers, scored with Bradley-Terry Elo.
It is closely followed by ElevenLabs' Eleven v4 Turbo at 1070 and Google DeepMind's Gemini 3.8 Flash-Lite TTS at 1037, with all three inside one statistical band, and holds clear statistical separation from the four models below.
Results backed by 590+ blind, head-to-head listener votes from native Indian English speakers.
See below for the exact audio raters judged and unedited listener verdicts 🧵
And the male voice, on a civic licence notice.
One sentence that carries a long alphanumeric licence number, a ward office, a room number and a deadline 🔊
What the raters typed after voting, unedited 👇
No self-reported benchmarks anywhere in this. Blind pairs, vetted native ears, Bradley-Terry math. That's the whole system.
Congrats to @KylanGibbs and the whole @inworld team 👏
Explore the board → voicearena.com/tts-leaderboa…
🏆 NVIDIA Nemotron 3.5 Transcribe debuts at #1 among 14 systems on Voice Arena’s Diarization + ASR leaderboard.
Across the full benchmark, it achieves 32.43% cpWER.
This track tests whether a system can get both the words and the speakers right. We evaluated 139 held-out conversations with 2-8 speakers across in-person and online settings, including overlapping speech.
Nemotron 3.5 Transcribe is first in both settings: 34.08% cpWER in person (far field) and 22.13% on online calls.
It has the lowest word deletion rate on the leaderboard. Deletions are 8.32% of reference words, against a median near 31%.
Congratulations to @NVIDIAAI on the release ! 👏
Congratulations to @Dheemanthredy and the @mayaresearch_ai team on releasing Maya-2-Native's weights under Apache 2.0: streaming TTS for 11 Indic languages and English.
On the Voice Arena Hindi TTS Leaderboard it ranks #2 of 9 among real-time models and #7 of 20 overall.
Maya-2 voice model weights are officially live! 🇮🇳🇮🇳
The frontier voice model for India is now completely open-weight.
• Outperforms ElevenLabs v3, Sarvam Bulbul v3, and Cartesia Sonic 3.5
• 1.5B+ people covered across all 28 states and 8 union territories
• SOTA voice synthesis across 11 official languages
• 20 native speakers pre-loaded, ready out of the box
• Best-in-class zero-shot voice cloning
Built by a team of five from the land of languages.
Go build and have fun!
Voice Arena retweeted
Maya-2 voice model weights are officially live! 🇮🇳🇮🇳
The frontier voice model for India is now completely open-weight.
• Outperforms ElevenLabs v3, Sarvam Bulbul v3, and Cartesia Sonic 3.5
• 1.5B+ people covered across all 28 states and 8 union territories
• SOTA voice synthesis across 11 official languages
• 20 native speakers pre-loaded, ready out of the box
• Best-in-class zero-shot voice cloning
Built by a team of five from the land of languages.
Go build and have fun!
Introducing five new languages on the Voice Arena TTS leaderboard.
Models from @ElevenLabs rank #1 on all five of the new Voice Arena TTS leaderboards: French 🇫🇷, German 🇩🇪, Bahasa 🇮🇩, Tagalog 🇵🇭 and Indian English 🇮🇳.
Eleven v4 Turbo takes the top position on four of the five boards. On Bahasa, ElevenLabs holds both #1 and #2, with Eleven v4 first and Eleven v4 Turbo second.
Across five languages spanning Europe, Southeast Asia and South Asia, each judged only by native listeners, ElevenLabs finishes first on every board, ahead of the next-best model from another provider by 11 to 26 Elo points.
Voice Arena ranks models through blind, head-to-head votes by vetted native speakers, scored with Bradley-Terry Elo.
Board-by-board results below 🧵
🇮🇩 Bahasa
ElevenLabs takes both of the top two places: Eleven v4 at #1 of 6 with 1118 Elo, and Eleven v4 Turbo at #2 with 1107.
voicearena.com/tts-leaderboa…
🚀 Mistral AI’s Voxtral Mini Transcribe 3 ranks #2 out of 9 models by DER on Voice Arena’s Diarization + ASR leaderboard, outperforming its larger counterpart.
Following the launch of Mistral’s new Voxtral 3 family, we evaluated its Mini and Small transcription models on Voice Arena’s Diarization Bench.
🥈 Voxtral Mini Transcribe 3: 40.4% DER (#2)
Voxtral Small Transcribe 3: 41.1% DER (#4)
Both models were tested on 139 held-out conversations under identical conditions, using human-verified references, a strict 0 ms collar, and no predefined speaker counts or audio cleanup.
Mini leads on in-person conversations (40.6% vs. 41.5%), while Small performs slightly better on online calls (39.1% vs. 39.5%).
Overlapping speech remains the biggest challenge. Missed speech accounts for 82% of Mini's total DER. Both models correctly identify the exact number of speakers in 83.5% of sessions.
Congratulations to @MistralAI on the release ! 👏
View complete track & condition breakdown: voicearena.com/diarization-b…
Murf AI's Falcon 2 moves from #11 to #2 of 14 on the Voice Arena US English 🇺🇸 latency board, with 102 ms to first audio at the median.
That is more than five times faster than its own earlier 528 ms, and puts Falcon 2 ahead of SpeechifyAI's Simba 3.2 at 123 ms and Inworld TTS 2 at 168.5 ms.
We measure time to first audio the way a listener hears it: leading silence counts, and no exact text is ever sent twice.
We're also evaluating Murf's latest Falcon 2 release in US English and Hindi, with results coming soon.
Congratulations to @MurfAIStudio on the update!
Explore the board → voicearena.com/tts-leaderboa…
Mistral AI's Voxtral-mini-tts-3-0 debuts at #10 of 23 on the Voice Arena US English 🇺🇸 TTS Leaderboard at 1011 Elo, ahead of Gradium-TTS, Fish Audio's Drama 3 and xAI's Grok TTS.
Voxtral-mini-tts-3-0 is the latest TTS model from @MistralAI, part of the Voxtral 3 release, with support for 35+ languages and 50+ dialects. Every ranking on Voice Arena comes from blind, head-to-head votes by vetted native speakers, scored with Bradley-Terry Elo.
It also ranks #6 of 20 on our Hindi board, making it one of seven models in the top 10 on both boards.
Results backed by 700+ blind, head-to-head listener votes from native US English speakers.
Congratulations to @MistralAI on the release!
See below for the exact audio raters judged and unedited listener verdicts 🧵
And the male voice, on a public service announcement.
A benefits notice for Cuyahoga County: a hard-to-pronounce place name, two acronyms and a batch code. SNAP is spoken as a word and EBT letter by letter, and a voice has to know which is which 🔊
What the raters typed after voting, unedited 👇
No self-reported benchmarks anywhere in this. Blind pairs, vetted native ears, Bradley-Terry math. That's the whole system.
Congrats to @arthurmensch and the whole @MistralAI team 👏
Explore the board → voicearena.com/tts-leaderboa…
