New Cartesia Text-to-Speech Model
Hi HN - team behind Cartesia here! We just announced our $27M seed today (https://techcrunch.com/2024/12/12/cartesia-claims-its-ai-is-...) + some major product updates we are excited to share with the developer community. Our latest releases: sonic-preview - a new model building on a new architecture that includes fundamental improvements over common TTS architectures which can struggle to follow complex, repetitive transcripts – you can find a fun example here (https://youtu.be/JTAmu8_qY4E)! Try this new model in the playground under…
In plain words
Cartesia offers a text-to-speech model called Sonic that converts written text into natural-sounding speech. The model uses a new architecture designed to handle complex and repetitive transcripts better than traditional TTS systems. It is built for developers who need speech generation capabilities and includes multilingual support with improved handling of numbers, dates, and times. The model is available through a playground and API.
written from the facts on this page · September 2026
From the sources
In the maker’s words, at launch
Hi HN - team behind Cartesia here! We just announced our $27M seed today (https://techcrunch.com/2024/12/12/cartesia-claims-its-ai-is-...) + some major product updates we are excited to share with the developer community. Our latest releases: sonic-preview - a new model building on a new architecture that includes fundamental improvements over common TTS architectures which can struggle to follow complex, repetitive transcripts – you can find a fun example here (https://youtu.be/JTAmu8_qY4E)! Try this new model in the playground under sonic-preview and it will be available via the API over the coming weeks. Multilingual Improvements - We launched a new multilingual model with enhanced number, date, and time recognition, as well as improved volume stability. Try it out in the playground or via our API! Why Developers Are Building With Cartesia Today: - We have the lowest latency at 90ms time to first audio, enabling truly real-time conversation - We have the highest quality, ultra-realistic voices as determined by respected third-party evals like Labelbox: https://labelbox.com/guides/evaluating-leading-text-to-speec.... - We can clone voices instantly, using only 5s of speech, and don’t cap the number of voices you can create - We can fine-tune professional voice clones using ~30 min of speech - We support 15 different languages, offer emotion and speed control, localization to different accents, and voice changer capabilities. - We offer all this and more at a more affordable price than competitors due to our innovative SSM architecture perfected for voice. SSMs offer clear advantages over transformers as they scale linearly with sequence length and enable cheap, high-throughput inference. Our founding team authored the widely cited Mamba SSM paper and our team has since built our models to be highly efficient, with better long-term memory, lower latency, and the ability to run locally on any device. - If you’re an early-stage startup building with voice, check out our startup grant (https://cartesia.ai/startup) or YC deal (https://deals.ycombinator.com/deals/2877) and qualify for 4 months on our scale tier (8M char/mo). You can try all of these features on our playground yourself: play.cartesia.ai, or if you have questions for the team or are interested in an enterprise plan with us, email [email protected].
More ai this month
the category →
I trained a 125M-parameter transformer to autocomplete piano performances in real time (~108 notes/sec on an iPhone 15). The idea is basically GitHub Copilot or Tabnine, except instead of prompting it with code, you prompt it by playing a few notes on a MIDI piano. The model then continues what you played, entirely on-device. The app is free if anyone wants to try it. Happy to answer questions about the model, training, Core ML, or the many things that didn't work.
AI · 17d ago · simedw.com
Astute▲585Automate your B2B brand going viral, with new media creators
AI · 18d ago · company-app.joinastute.com
Hey HN, Henry from Cactus here! We previously released Cactus Needle, a 14MB agentic LLM for tool call, device use, and structured extraction for phones, wearables, smart homes, small robots and microcontrollers. We got really great feedback here, and have now incorporated the suggestions to release Needle 2. The whole model is a single 14MB binary that runs a full session in 28MB of RAM; 45m parameters at 2bit compression. Needle hits 500 tokens/sec decode speed on a Raspberry Pi 5, sits between 400-1,500 tokens/sec on VR devices like Meta Quest 3S and Apple Vision Pro, and ranges…
AI · 26d ago · cactuscompute.com
