Releases·Model Releasefeatured
Meta Superintelligence Labs releases Muse Voice Transcribe, first real-time audio perception model
Meta Superintelligence Labs launched Muse Voice Transcribe, a real-time streaming ASR model with diarization for 20+ speakers and endpointing. It is an autoregressive multimodal LLM from the Muse Spark family that processes audio in 80ms chunks and decides per chunk whether to keep listening or emit text, trained with RL on word error rate and delay rewards. It is available via the Meta Model API, Meta AI for Mac, and Muse Code.
1384 engagement3 sourcesTue, Sep 1, 2026, 05:26 PM
AI Overview
- Processes audio in 80ms chunks (12.5 Hz), one token per chunk, deciding at each step to keep listening or emit text
- Trained with reinforcement learning using combined word error rate and delay rewards
- Supports multilingual transcription with seamless code-switching
- Positioned as the first real-time audio perception model from Meta Superintelligence Labs
- Ships via Meta Model API, Meta AI for Mac, and Muse Code
Related
Releases·Model Release
MeanVC2 streaming voice conversion model released with cross-gender, cross-language support
ASLP-lab released MeanVC2, a streaming voice conversion model that supports cross-gender and cross-language conversion. It runs 3x realtime on CPU via audio.cpp, enabling real-time voice conversion without dedicated hardware.
ASLP-lab released MeanVC2, a streaming voice conversion model that supports cross-gender and cross-language conversion. It runs 3x realtime on CPU via audio.cpp, enabling real-time voice conversion without dedicated hardware.
50 engagement1 sourceredditFri, Aug 21, 2026, 02:03 AM
Releases·Model Release
NVIDIA releases Nemotron 3 Diarization for speaker diarization on batch and streaming audio
NVIDIA released Nemotron 3 Diarization, a model for speaker diarization (identifying who spoke when) supporting both batch and streaming processing, announced via a Hugging Face blog and weights on Hugging Face. It reportedly outperforms many other diarization models and is evaluated with DER (diarization error rate). A YouTube walkthrough demonstrates it running on DGX Spark with NeMo, including full-podcast transcription exportable as text or SRT — relevant to developers building voice agents.
NVIDIA released Nemotron 3 Diarization, a model for speaker diarization (identifying who spoke when) supporting both batch and streaming processing, announced via a Hugging Face blog and weights on Hugging Face. It reportedly outperforms many other diarization models and is evaluated with DER (diarization error rate). A YouTube walkthrough demonstrates it running on DGX Spark with NeMo, including full-podcast transcription exportable as text or SRT — relevant to developers building voice agents.
657 engagement1 sourceyoutubeWed, Sep 23, 2026, 03:36 PM
Products·building
Real-time voice-to-voice translator preserving voice, tone, and emotions
VertoX is a real-time voice-to-voice translation platform that preserves the speaker's original voice, tone, and emotions. It uses LLMs for natural language understanding and generation, enabling seamless multilingual conversations without text intermediaries. The product solves the problem of losing personal expression in translation for travelers, business professionals, and students.
VertoX is a real-time voice-to-voice translation platform that preserves the speaker's original voice, tone, and emotions. It uses LLMs for natural language understanding and generation, enabling seamless multilingual conversations without text intermediaries. The product solves the problem of losing personal expression in translation for travelers, business professionals, and students.
18 engagement1 sourceredditSun, Jul 19, 2026, 02:13 PM
Developments·Paper
MEUSLI: open-source multilingual projector family for Whisper + LLM ASR in 28 European languages
Researchers introduce MEUSLI, the first open-science multilingual projector family that connects a Whisper speech encoder with open-source multilingual LLMs, enabling fully open-source end-to-end ASR in 28 European languages. The projectors map acoustic features into token-level embeddings, extending prior monolingual pipelines to achieve strong results across diverse languages.
Researchers introduce MEUSLI, the first open-science multilingual projector family that connects a Whisper speech encoder with open-source multilingual LLMs, enabling fully open-source end-to-end ASR in 28 European languages. The projectors map acoustic features into token-level embeddings, extending prior monolingual pipelines to achieve strong results across diverse languages.
0 engagement1 sourcearxivFri, Jul 24, 2026, 08:52 AM
Releases·Tool Release
Muse announces adaptive delay for Muse Voice Transcribe, claiming pareto frontier on speed-accuracy trade-off
Muse announced that its Voice Transcribe product uses an adaptive delay mechanism to reach the pareto frontier on the speed-accuracy trade-off, measured by time to final transcription. The claim suggests the system dynamically balances how long it waits before finalizing a transcript against accuracy. Practitioners evaluating streaming transcription tools may find the latency-accuracy framing relevant.
Muse announced that its Voice Transcribe product uses an adaptive delay mechanism to reach the pareto frontier on the speed-accuracy trade-off, measured by time to final transcription. The claim suggests the system dynamically balances how long it waits before finalizing a transcript against accuracy. Practitioners evaluating streaming transcription tools may find the latency-accuracy framing relevant.
49 engagement1 sourcexTue, Sep 1, 2026, 05:26 PM