Amu

Building Canopy Labs where we make virtual humans move and speak.

[Amu @akv_13]

@Devon07979992 What milestone would you say this hit?

[Amu @akv_13]

Voice most important modality after text.

Speech-LLMs are the future.

Here is a chronological list of the important Open Source milestones in the Speech-LLM space 📖.

A thread 🧵 ...

[Amu @akv_13]

@dctanner @uselayercode Awesome! Excited to see it!

[Amu @akv_13]

@kadirnardev was trying to decide between spark and llasa :)

[Amu @akv_13]

Qwen 2.5 Omni (03/2025)

“No catastrophic forgetting in e2e speech”

https://t.co/wQuaZn9N22

[Amu @akv_13]

Orpheus (03/2025)

“Finally OS speech >= closed source”

https://t.co/MJCejfCiJH

[Elias @Eliasfiz]

We also sponsor visas!

Apply by emailing us: jobs[at]canopylabs[dot]ai

[Thomas Wolf @Thom_Wolf]

Orpheus a SOTA text-to-speech model
https://t.co/yLObMCEpS2

[Elias @Eliasfiz]

OrpheusTTS is now at 200k downloads on Hugging Face with 4.7k stars and 400 forks on Github!

The OS community has been crushing it

[Amu @akv_13]

@m4zas24 Hi - we'd appreciate any suggestions - please feel free to post them here/dm me/ or put them in a github discussion!

[Amu @akv_13]

We designed Orpheus to be easily fine tuned. On Hugging Face 🤗 there are over 200 finetunes of our model. People have made Orpheus:

- Speak dozens of languages 🌎
- Create personalised voice interfaces 🔊
- Recreate historical/fictional characters📜

[Elias @Eliasfiz]

Today, we’re launching Orpheus, an open-source TTS model that exceeds the capabilities of both open and closed-source models such as ElevenLabs and OpenAI! (1/6)

[Philip Kiely @philipkiely]

Deploying and vibe checking Orpheus TTS, an open-source model for generating speech.

Our implementation supports up to 48 concurrent real-time users per H100 GPU!

[Amu @akv_13]

Excited to bring optimised inference for Orpheus to everyone!

[Elias @Eliasfiz]

People told us they want Orpheus TTS in production.

So we partnered with @baseten as our preferred inference provider!

Baseten runs Orpheus with:

•⁠ ⁠Low latency (<200 ms TTFB)
•⁠ ⁠High throughput (up to 48 real-time streams per H100)
•⁠ ⁠Secure, worldwide infra

[Jack @jackndwyer]

Building an AI app with $1/hr AI voice, memory, tool-calling, phone—that's what you use Gabber for

But time to value shouldn't be hours, it should be seconds

Build a sample AI companion in 20 seconds, try it, tune it, then go live with it (last part coming soon)

[SebastianBoo @SebastianB929]

Thanks for the OrpheusTTS release!

Great models, easy to finetune and you can even livestream using vllm or sglang with a single RTX 3090 with FP8 quantization. INT4 should be even faster.😃

[Amu @akv_13]

ps: we're hiring https://t.co/6UniFMSjfW

[Amu @akv_13]

Orpheus has hit it's first 100k downloads on HuggingFace 🤗. A first small step, with many strides to follow shortly, starting with Orpheus v1!