ASR

OpenAI has released Whisper large-v3, an open-weight automatic speech recognition and translation model trained on 5 million hours of audio. The model reduces transcription errors by 10-20% compared to its predecessor and adds native Cantonese support.

Microsoft Research released VibeVoice-ASR-BitNet, a quantized 1.58GB version of its VibeVoice-ASR speech recognition model that achieves real-time inference (RTF < 1) on as few as 3 CPU threads. The model runs 1.6-2.3x faster than Whisper.cpp on commodity x86 and ARM hardware, with a modest accuracy tradeoff.

ServiceNow released AU-Harness, the first comprehensive benchmark for code-switched speech recognition in enterprise voice agents, testing seven ASR systems including ElevenLabs, Gemini, and AssemblyAI. The benchmark covers 918 utterances across Spanish-English, French-English, Canadian French-English, and German-English, measuring Word Error Rate (WER), Semantic WER (SWER), and Answer Error Rate (AER). ElevenLabs Scribe V2 achieved the lowest WER across all language pairs, followed closely by AssemblyAI Universal-3 Pro.

NVIDIA released Nemotron 3.5 ASR, a 600M-parameter speech-to-text model supporting 40 language-locales from a single checkpoint. The model achieves 0.07 seconds to final transcript after speech ends and ranks 2nd in latency among streaming ASR models according to Artificial Analysis benchmarks.

IBM has released Granite Speech 4.1 2B, a 2-billion-parameter speech-language model trained on 174,000 hours of audio for automatic speech recognition and translation across English, French, German, Spanish, Portuguese, and Japanese. The model introduces a dual-head CTC encoder and includes variants for speaker attribution and a novel non-autoregressive architecture for higher throughput.