正在加载视频...

视频加载失败

加载此视频时出现问题。这可能是由于临时网络问题，或视频可能不可用。

introducing rumik oss 1 india's first state of the art open source, fully pre-trained voice model capable of speaking 20+ languages trained on only 66k hours of data, it beats even the top closed models on multiple benchmarks including emotion realism weights are now live 👇🏻

rohan

4,889 subscribers

100,771 次观看 • 12 天前 •via X (Twitter)

Anya Rossi• Live Now

Private livecam show

47 条评论

rohan 的头像

rohan12 天前

rumik oss 1 is a 3 billion parameter open weights text to speech model it can speak in indian english, hindi, telugu, tamil, kannada, bengali, punjabi and switches between them mid-sentence, and takes delivery instructions in plain text full research blog - download the weights on hugging face -

rohan 的头像

rohan12 天前

describe how you want the model to sound like, and it will follow those instructions here are some examples 👇🏻

rohan 的头像

rohan12 天前

along with the model, we are also open sourcing two of our internal benchmarks 1. indic-emo this benchmark tests can a model hold an emotion through code-switched speech? 100 prompts, each mixing two to five of our seven languages, equal coverage of happy, sad, angry, excited and professional. three judge models score anonymized clips on a five-point rubric covering tone, dynamics, phrasing and sustained expression. rumik oss 1 comfortably beats @cartesia, @ElevenLabs and only @GeminiApp is better and by a big margin it was expected given gemini 3.1 flash tts must be trained on 1000x more data compared to rumik oss 1 is only trained on 66k hours in total

rohan 的头像

rohan12 天前

2. nova this benchmark measures precise delivery of emotions for eg - if you ask for a laugh at word seven, does it land at word seven? 149 english prompts, one vocalization each at a specified word position, seven systems, three synthesis runs. three detectors check position-correct rendering. benchmark + protocol ->

rohan 的头像

rohan12 天前

on WER (word error rate) and CER (character error rate) rumik oss 1 tops odia, konkani, maithili while comes second in assamese, kannada, nepali, malyalam

rohan 的头像

rohan12 天前

rumik oss 1 comes with CC BY-NC 4.0 license plus an acceptable use addendum. that makes it open weights for research and non-commercial use, speech research, benchmarking, teaching, finetuning for experiments. you can share the weights and your adaptations under the same terms, keeping attribution, license links and a note of what you changed.

Soumith Chintala 的头像

Soumith Chintala12 天前

congrats to the @rumik_ai team. looks like a good first attempt. hope to see rapid iterations.

rohan 的头像

rohan12 天前

@rumik_ai thanks a ton @soumithchintala

Arpan (Locked In ARC) 的头像

Arpan (Locked In ARC)12 天前

LFG🔥🔥

Arpan (Locked In ARC) 的头像

Arpan (Locked In ARC)12 天前

You will win 🍀

Vaibhav Srivastava 的头像

Vaibhav Srivastava12 天前

Congratulations 🎊

Rahul Guha 的头像

Rahul Guha12 天前

I was using @SarvamAI, let's see how this plays out.

rohan 的头像

rohan12 天前

@SarvamAI let's go

Mad scientist 🥼 的头像

Mad scientist 🥼12 天前

Congratulations, how can we contribute?

rohan 的头像

rohan12 天前

Spread the word 🤌

AMBUJ TIWARI 的头像

AMBUJ TIWARI12 天前

I'll make sure to cover each update from now 🙌

rohan 的头像

rohan12 天前

hahaha indeed

cheaty 的头像

cheaty12 天前

took you guys long enough

∞ Gman 的头像

∞ Gman12 天前

Great move to open source it 💪 We should add them to @indos_tech catalog @utkarsh_k1c

Genysis AI 的头像

Genysis AI12 天前

Great I am happy india is taking root in ai landscape

Samiksha Shukla 的头像

Samiksha Shukla12 天前

Crazy! Excited to try it out!!

rohan 的头像

rohan12 天前

lmk how it goes!

pushkar 的头像

pushkar12 天前

super cool, congratulations on the release!

rohan 的头像

rohan12 天前

thanks @thepushkarp

Mehar 的头像

Mehar12 天前

never not impressed by rumik. so so good 🙌

Brian Hadu 的头像

Brian Hadu12 天前

if it really beats top closed models, it could shift the landscape for accessibility

himanshu 的头像

himanshu12 天前

very cool, congrats!

rohan 的头像

rohan11 天前

tahnks @himanshustwts

Aryan 的头像

Aryan12 天前

🚀🚀🚀🚀

Kimi Nguyen 的头像

Kimi Nguyen12 天前

training on just 66k hours and outperforming top models is no small feat

Ayman 的头像

Ayman12 天前

Congratulations 🎉. Is Bengali trained as well?

rohan 的头像

rohan12 天前

Yes

Parth 的头像

Parth12 天前

very cool, congrats on the launch!

aniketh 的头像

aniketh12 天前

never bet against these guys @lets_dig_deeper and @bhartivatsal ❤️

Srinivasa Babu 💙 的头像

Srinivasa Babu 💙12 天前

It's exciting to know that we have multiple options across different languages, which can now be used as an alternative to Google's live models. Though at this point in time, I would still prefer using Gemini Live because I feel I can make it sound better in Tamil and have the same expressiveness that has been demonstrated here. I am really excited to try this on the next version, though!

Noctus 的头像

Noctus12 天前

Congrats for the launch

ᱫᱩᱞᱟᱹᱲ 的头像

ᱫᱩᱞᱟᱹᱲ12 天前

does it support Santali?

kartik 的头像

kartik12 天前

insane🔥, and well detailed blog!

Aru_sharma 的头像

Aru_sharma12 天前

Let's go congrats for the crazy feat

Krishna Chaturvedi 的头像

Krishna Chaturvedi12 天前

66k hours and these benchmark results is seriously impressive. Great work.🤌

Atharva 的头像

Atharva12 天前

Incredible timing, I just spent half a day messing with Sarvam to make it sound right. Congrats!

Arib Zeeshan 的头像

Arib Zeeshan12 天前

This is the first indian AI thing that is actually superior to the other peer products.........I dont know how well it is compared to GPT Live (Its the current SOTA I think), But certainly one of the better open source models from the demos I have seen.

Hrushik Pabbathi 的头像

Hrushik Pabbathi12 天前

yooo that's insane!! congrats!!

Neelakandan NC 的头像

Neelakandan NC12 天前

Languages are finally becoming our strength

Nipun Taneja 的头像

Nipun Taneja12 天前

W

Rohan Malik 的头像

Rohan Malik12 天前

from Rohan to another....keep at it bro...naam roshan karo humara.

Vineeth 的头像

Vineeth12 天前

Lfggggg

相关视频

The new open-source Text to Speech model: Fish Speech 1.4 is brilliant! Trained on a massive 700K hours of multilingual speech data in 8 languages - Instant voice cloning 🗣️ - Ultra-low latency ⚡ - Compact model (~1GB weights) 🏋️‍♂️

The new open-source Text to Speech model: Fish Speech 1.4 is brilliant! Trained on a massive 700K hours of multilingual speech data in 8 languages - Instant voice cloning 🗣️ - Ultra-low latency ⚡ - Compact model (~1GB weights) 🏋️‍♂️

Rohan Paul

228,905 次观看 • 2 年前

Introducing Indic-Parler TTS - Trained on 10K hours of data, 938M params, supports 20 Indic languages, emotional synthesis, apache 2.0 licensed! 🔥 A collaboration w/ AI4Bharat & Hugging Face - w/ fully customisable speech and voice personas! Try it out directly below or use the model weights as you want! 🇮🇳/acc

Introducing Indic-Parler TTS - Trained on 10K hours of data, 938M params, supports 20 Indic languages, emotional synthesis, apache 2.0 licensed! 🔥 A collaboration w/ AI4Bharat & Hugging Face - w/ fully customisable speech and voice personas! Try it out directly below or use the model weights as you want! 🇮🇳/acc

Vaibhav (VB) Srivastav

42,210 次观看 • 1 年前

Introducing Ψ₀ ( — an open foundation model for universal humanoid loco-manipulation. 🏆 Outperforms GR00T N1.6 by 40%+ overall success rate 📉 Uses only ~10% of the pre-training data 📦 Fully open-source: model, data, code, and deployment pipeline 1/10

Introducing Ψ₀ ( — an open foundation model for universal humanoid loco-manipulation. 🏆 Outperforms GR00T N1.6 by 40%+ overall success rate 📉 Uses only ~10% of the pre-training data 📦 Fully open-source: model, data, code, and deployment pipeline 1/10

Yue Wang

26,408 次观看 • 6 个月前

LETS GOO! Parler TTS 🔥 A fully open-source, Apache 2.0 licensed Text-to-speech model focused on providing maximum controllability. Through voice prompts, you can control the pitch, speed, gender, noise levels, emotion characteristics and more! > Trained on 10K hours of permissive data. > Offers control over the generations. > Training + Inference code released. > The processed dataset and tagging scripts were released for further research. > English only for now. Next, we're scaling the training to 50K hours and even better dataset processing! Want to help us out? DMs open! 🤗

LETS GOO! Parler TTS 🔥 A fully open-source, Apache 2.0 licensed Text-to-speech model focused on providing maximum controllability. Through voice prompts, you can control the pitch, speed, gender, noise levels, emotion characteristics and more! > Trained on 10K hours of permissive data. > Offers control over the generations. > Training + Inference code released. > The processed dataset and tagging scripts were released for further research. > English only for now. Next, we're scaling the training to 50K hours and even better dataset processing! Want to help us out? DMs open! 🤗

Vaibhav (VB) Srivastav

156,386 次观看 • 2 年前

1/n We are excited to announce EchoPrime – the first echocardiography AI model capable of evaluating a full transthoracic echocardiogram study, identify the most relevant videos, and produce a comprehensive interpretation! Great work lead by Milos Vukadinovic, EchoPrime is the largest and most complex #echofirst AI model yet, trained on 10x the data of existing models.

1/n We are excited to announce EchoPrime – the first echocardiography AI model capable of evaluating a full transthoracic echocardiogram study, identify the most relevant videos, and produce a comprehensive interpretation! Great work lead by Milos Vukadinovic, EchoPrime is the largest and most complex #echofirst AI model yet, trained on 10x the data of existing models.

David Ouyang, MD

155,571 次观看 • 1 年前

LETS GOOO! FishSpeech v1.5 - multilingual, zero-shot instant voice cloning, low-latency, open text to speech model 🔥 > Only 500M params > Trained on 1 MILLION hours of audio > Supports 13 languages > Low-latency ( Open model - checkpoints on the hub 🤗 > Best part: Ranks #2 on TTS Arena (preliminary results) Kudos to Fish Audio - truly leading the way for open text to speech research! ⚡

LETS GOOO! FishSpeech v1.5 - multilingual, zero-shot instant voice cloning, low-latency, open text to speech model 🔥 > Only 500M params > Trained on 1 MILLION hours of audio > Supports 13 languages > Low-latency ( Open model - checkpoints on the hub 🤗 > Best part: Ranks #2 on TTS Arena (preliminary results) Kudos to Fish Audio - truly leading the way for open text to speech research! ⚡

Vaibhav (VB) Srivastav

49,150 次观看 • 1 年前

Today we are introducing Dyna-2, a world-action model pre-trained on one million hours of human video. At this scale, for the first time, we discovered several new scaling laws: • world-action models exhibit scaling law on human data across four orders of magnitude, from 1000 to 1,000,000 hours, • this human data scaling law implied a scaling law on never seen robot data, • both data and objective matter; world modeling and scaling on video data are essential for cross-embodiment scaling transfer to emerge 🧵

Today we are introducing Dyna-2, a world-action model pre-trained on one million hours of human video. At this scale, for the first time, we discovered several new scaling laws: • world-action models exhibit scaling law on human data across four orders of magnitude, from 1000 to 1,000,000 hours, • this human data scaling law implied a scaling law on never seen robot data, • both data and objective matter; world modeling and scaling on video data are essential for cross-embodiment scaling transfer to emerge 🧵

Dyna Robotics

1,794,709 次观看 • 1 个月前

Introducing TranslatePsy-AfriSLM: state-of-the-art machine translation for 19 Sub-Saharan African languages using models small enough to run on a laptop or a smartphone. Our smallest LM, at just 0.8B parameters, matches or beats systems up to 152 times its size, including Qwen3.5-122B-A10B and TranslateGemma-27B, on 3 key benchmarks. Models: Research Paper: Blog:

Introducing TranslatePsy-AfriSLM: state-of-the-art machine translation for 19 Sub-Saharan African languages using models small enough to run on a laptop or a smartphone. Our smallest LM, at just 0.8B parameters, matches or beats systems up to 152 times its size, including Qwen3.5-122B-A10B and TranslateGemma-27B, on 3 key benchmarks. Models: Research Paper: Blog:

QVAC

386,787 次观看 • 18 天前

Introducing GLM 5.2 for autoresearch GLM 5.2 is the first open weights model we've tried on our autoresearch pipeline that's proven capable for real research tasks. With Fable 5's restrictions on research, having an open weights alternative is a huge win for open source Watch it carry out fully async vs colocated sync RL training on Harbor code contests across two 8xH100 nodes on top of SkyRL. Resolves setup issues, tracks runs to completion, and produces a full comparison of throughput and reward stability

Introducing GLM 5.2 for autoresearch GLM 5.2 is the first open weights model we've tried on our autoresearch pipeline that's proven capable for real research tasks. With Fable 5's restrictions on research, having an open weights alternative is a huge win for open source Watch it carry out fully async vs colocated sync RL training on Harbor code contests across two 8xH100 nodes on top of SkyRL. Resolves setup issues, tracks runs to completion, and produces a full comparison of throughput and reward stability

alphaXiv

271,866 次观看 • 3 个月前

Introducing Scribe — the most accurate Speech to Text model. It has the highest accuracy on benchmarks, outperforming previous state-of-the-art models such as Gemini 2.0 and OpenAI Whisper v3. It’s now the leading model for English, Spanish, Italian, and many more. With support for 99 languages, speaker diarization, character-level timestamps, and non-speech events such as laughing.

Introducing Scribe — the most accurate Speech to Text model. It has the highest accuracy on benchmarks, outperforming previous state-of-the-art models such as Gemini 2.0 and OpenAI Whisper v3. It’s now the leading model for English, Spanish, Italian, and many more. With support for 99 languages, speaker diarization, character-level timestamps, and non-speech events such as laughing.

ElevenLabs

464,967 次观看 • 1 年前

Mfw Minimax H3, the new Open Source Audiovisual model that is vastly superior to all the Frontier models, is also the first model in history to be trained on and remember The Will Stancil Show and now chooses to generate propaganda posters of Will Stancil all on its own.

Mfw Minimax H3, the new Open Source Audiovisual model that is vastly superior to all the Frontier models, is also the first model in history to be trained on and remember The Will Stancil Show and now chooses to generate propaganda posters of Will Stancil all on its own.

Emily Youcis

13,539 次观看 • 1 个月前

#WATCH | Bengaluru, Karnataka: Co-founder, Sarvam AI, Pratyush Kumar says, "... It's been three years now. We started off building on top of the Llama model, which was the best open model at that time. Sarvam was the only company outside, or maybe in Asia, that contributed to the Llama 3 Series. We understood how to add language skills into models, and Llama 3 was reasonable in Indian languages. When the first set of GPUs arrived in India, we pre-trained from scratch a 2-billion model and worked with Infosys to make it useful for IT services. Then we recognised the importance of scaling up reinforcement learning and built a reasoning model called Sarvam M on top of Mistral's small model. Over the last year, we pre-trained from scratch in India on 4,000 Hopper Series GPUs a 100-billion-plus-parameter model..."

#WATCH | Bengaluru, Karnataka: Co-founder, Sarvam AI, Pratyush Kumar says, "... It's been three years now. We started off building on top of the Llama model, which was the best open model at that time. Sarvam was the only company outside, or maybe in Asia, that contributed to the Llama 3 Series. We understood how to add language skills into models, and Llama 3 was reasonable in Indian languages. When the first set of GPUs arrived in India, we pre-trained from scratch a 2-billion model and worked with Infosys to make it useful for IT services. Then we recognised the importance of scaling up reinforcement learning and built a reasoning model called Sarvam M on top of Mistral's small model. Over the last year, we pre-trained from scratch in India on 4,000 Hopper Series GPUs a 100-billion-plus-parameter model..."

ANI

163,060 次观看 • 1 个月前

Introducing DINOv3: a state-of-the-art computer vision model trained with self-supervised learning (SSL) that produces powerful, high-resolution image features. For the first time, a single frozen vision backbone outperforms specialized solutions on multiple long-standing dense prediction tasks. Learn more about DINOv3 here:

Introducing DINOv3: a state-of-the-art computer vision model trained with self-supervised learning (SSL) that produces powerful, high-resolution image features. For the first time, a single frozen vision backbone outperforms specialized solutions on multiple long-standing dense prediction tasks. Learn more about DINOv3 here:

AI at Meta

901,265 次观看 • 1 年前

🚀 Galbot just released AstraBrain-WBC 0.5, the world’s first humanoid robot cerebellum foundation model. Trained on 20,000 hours of human motion data, it validates a Scaling Law for robot motion control and demonstrates zero-shot execution of unseen motions on real humanoids. The era of robot control foundation models is getting closer.

🚀 Galbot just released AstraBrain-WBC 0.5, the world’s first humanoid robot cerebellum foundation model. Trained on 20,000 hours of human motion data, it validates a Scaling Law for robot motion control and demonstrates zero-shot execution of unseen motions on real humanoids. The era of robot control foundation models is getting closer.

Eren Chen

10,270 次观看 • 3 个月前

NEW: Higgs Audio V2 from BosonAI open, unified TTS model w/ voice cloning, beats GPT 4o mini tts and ElevenLabs v2 🔥 > Trained on 10M hours (speech, music, events) > Built on top of Llama 3.2 3B > Works real-time and on edge > Beats GPT-4o-mini-tts, ElevenLabs v2 in prosody & emotion Multi-speaker dialog > Zero-shot voice cloning 🤩 > Available on Hugging Face Kudos to folks at Boson AI for releasing such a brilliant work and all the details around the model! 🤗

NEW: Higgs Audio V2 from BosonAI open, unified TTS model w/ voice cloning, beats GPT 4o mini tts and ElevenLabs v2 🔥 > Trained on 10M hours (speech, music, events) > Built on top of Llama 3.2 3B > Works real-time and on edge > Beats GPT-4o-mini-tts, ElevenLabs v2 in prosody & emotion Multi-speaker dialog > Zero-shot voice cloning 🤩 > Available on Hugging Face Kudos to folks at Boson AI for releasing such a brilliant work and all the details around the model! 🤗

Vaibhav (VB) Srivastav

79,645 次观看 • 1 年前

- <1B params - supports 91 languages - 5 pages/s on RTX 5090 - runs on CPU, GPU, MPS - 83.3% olmocr bench score (top under 3B) Surya OCR is a state-of-the-art model for document intelligence. 100% open-source.

- <1B params - supports 91 languages - 5 pages/s on RTX 5090 - runs on CPU, GPU, MPS - 83.3% olmocr bench score (top under 3B) Surya OCR is a state-of-the-art model for document intelligence. 100% open-source.

Akshay 🚀

63,902 次观看 • 3 个月前