FireRedAudio
One open 9B audio model for speech recognition, understanding, TTS and editing
Emerged finds new products worth a look, daily — what’s new today
VoiceMem
Open streaming memory for real-time voice agents, split into facts and feelings
If you are building a voice agent, this is an open, benchmarked way to give it long-term memory without slowing every turn.
VoiceMem is memory infrastructure for voice agents. A dual-brain design keeps facts about the user in one store and emotional context in another, streaming the relevant pieces to the model during a live conversation. The authors report 134ms retrieval against roughly 1.4s for comparable systems. The release includes code, evaluation, a fine-tuning pipeline, model weights on Hugging Face, the ChatMem-400K dataset and an arXiv report, under Apache-2.0.
More like this
Nari Qwen3-TTS and Qwen3-ASR
Fast, low-cost speech APIs built on Qwen3 voice models
DaiDocs
An open plain-text file format for AI memory that any model can read
Fraise
A single-binary memory database for AI agents with a two-verb query language
Egma
Open-source simulation testing and monitoring for voice agents
Sparrow-2
A turn-taking model that knows when a voice agent should speak