Toby Kim

Two undergrads. One still in the military. Zero funding. One ridiculous goal: build a TTS model that rivals NotebookLM Podcast, ElevenLabs Studio, and Sesame CSM. Somehow… we pulled it off. Here’s how 👇

@fabianstelzer Valid point. But there are so many great models on Hugging Face that are being used thousands of times a day - which aren’t from top 10 labs. I just think consumer adoption is lagging behind

@_doyeob_ correct me if I’m wrong but there is very little demand for non top 10 foundation models so I’m not sure the metaphor makes sense?

Model Training is the new Building Apps. 15 years ago, building a mobile app was hard. But a small technical team could pull it off. Same for model training nowadays. Only blocker is compute. But programs like TRC and other grants are fixing that. Excited for the future of AI.

@NotebookLM Exciting times for audio!

This just in... the @NotebookLM hosts have some rather exciting news they'd like to share with you all:

@nirajshah Thank you for the kind words! We’re going to keep on building and building :)))

@_doyeob_ Great stuff.

@Sathees89347227 We’re currently working with the @huggingface team to bring Dia to transformers for training / finetune / fast inference.

@_doyeob_ We need training & fine tuning scripts🙂

@anonomize we're going to ship something much greater than a GUI

@_doyeob_ We need a GUI

@PyMan_Official we're working on it + have some good leads

@_doyeob_ speech is too fast, fix that also

Nari Waitlist: tally.so/r/meokbo

Dia hit +6.5k stars on Github, #1 trending on Hugging Face - in under 48 hours. Thank you for all the support. But we won't stop here. We are building a product to change the future of audio entertainment. Curious? Join the waitlist 👇

Try it now on Github: github.com/nari-labs/dia HF: huggingface.co/spaces/nari-la… Waitlist for hosted version with improved models: tally.so/r/meokbo

+3.4k stars on Github. #2 trending on Hugging Face. All in under 24 hours. Thanks for all the support <3 Here's Dia speaking about our launch:

2k stars in just over 12 hours! thanks for the huge support <3 (also top 5 on huggingface trending!) will be delivering big upgrades to the repo throughout this week.

People seemed to be enjoying the fire example, so I'm also adding it to the thread here. Thanks for all the support :)))

Try Dia now! Github: github.com/nari-labs/dia/ HF: huggingface.co/nari-labs/Dia-…

Try Dia now! Github: github.com/nari-labs/dia/ HF: huggingface.co/spaces/nari-la… Waitlist for hosted version with larger models: tally.so/forms/meokbo

A fun comparison between Dia-1.6B versus Sesame's CSM-1B on emotional conversations. (contains strong language) 👑Who won?

🎧 Dia is a 1.6B parameter text to speech model created by Nari Labs. (Apache 2.0) 🔊 Jupyter Notebook 🥳 Thanks to @_doyeob_ ❤ Nari Labs ❤ 🌐page: tally.so/r/meokbo 🧬code: github.com/nari-labs/dia 🍊jupyte: github.com/camenduru/dia-…

3 months later, we had a fully trained 1.6B model. It took longer and was harder than expected, but totally worth it. The best time to start is today!

DeepMind’s How to Scale and HuggingFace’s Ultra-Scale Playbook were super helpful. If you are interested in training large models, go read them now!

But we wanted more. More control over the voices. More freedom in the script. We tried every TTS API on the market — none of them sounded like real human conversation.