Audio Language Model

Can a speech enhancer educated solely on actual noisy recordings cleanly separate speech and noise—with out ever seeing paired information? A crew of researchers from Brno University of Technology and Johns Hopkins University proposes Unsupervised Speech Enhancement utilizing Data-defined Priors (USE-DDP), a dual-stream encoder–decoder that separates any noisy enter into two waveforms—estimated clear speech and…

Liquid AI has launched LFM2-Audio-1.5B, a compact audio–language basis mannequin that each understands and generates speech and textual content via a single end-to-end stack. It positions itself for low-latency, real-time assistants on resource-constrained units, extending the LFM2 household into audio whereas retaining a small footprint. https://www.liquid.ai/weblog/lfm2-audio-an-end-to-end-audio-foundation-model But what’s really new? a unified spine with disentangled…

Xiaomi’s MiMo crew launched MiMo-Audio, a 7-billion-parameter audio-language mannequin that runs a single next-token goal over interleaved textual content and discretized speech, scaling pretraining past 100 million hours of audio. What’s really new? Instead of relying on task-specific heads or lossy acoustic tokens, MiMo-Audio makes use of a bespoke RVQ (residual vector quantization) tokenizer that…

TwinMind, a California-based Voice AI startup, unveiled Ear-3 speech-recognition mannequin, claiming state-of-the-art efficiency on a number of key metrics and expanded multilingual help. The launch positions Ear-3 as a aggressive providing towards present ASR (Automatic Speech Recognition) options from suppliers like Deepgram, AssemblyAI, Eleven Labs, Otter, Speechmatics, and OpenAI. Key Metrics Metric TwinMind Ear-3 Result…

Alibaba Cloud’s Qwen workforce unveiled Qwen3-ASR Flash, an all-in-one automated speech recognition (ASR) mannequin (obtainable as API service) constructed upon the sturdy intelligence of Qwen3-Omni that simplifies multilingual, noisy, and domain-specific transcription with out juggling a number of techniques. Key Capabilities Multilingual recognition: Supports automated detection and transcription throughout 11 languages together with English and…

The Allen Institute for AI (AI2) has launched OLMoASR, a set of open automated speech recognition (ASR) fashions that rival closed-source programs comparable to OpenAI’s Whisper. Beyond simply releasing mannequin weights, AI2 has revealed coaching information identifiers, filtering steps, coaching recipes, and benchmark scripts—an unusually clear transfer in the ASR area. This makes OLMoASR one…

Introduction Tencent’s Hunyuan staff has launched Hunyuan-MT-7B (a translation mannequin) and Hunyuan-MT-Chimera-7B (an ensemble mannequin). Each fashions are designed particularly for multilingual machine translation and had been launched along side Tencent’s participation within the WMT2025 Normal Machine Translation shared activity, the place Hunyuan-MT-7B ranked first in 30 out of 31 language pairs. https://github.com/Tencent-Hunyuan/Hunyuan-MT/blob/foremost/Hunyuan_MT_Technical_Report.pdf Mannequin Overview…

Microsoft AI lab formally launched MAI-Voice-1 and MAI-1-preview, marking a brand new section for the corporate’s synthetic intelligence analysis and improvement efforts. The announcement explains how Microsoft AI Lab is getting concerned in AI analysis with none third social gathering involvement. MAI-Voice-1 and MAI-1-preview fashions helps distinct however complementary roles in speech synthesis and general-purpose…