PulseAugur
EN
LIVE 13:37:25
Русский(RU) Cohere открыла код модели распознавания арабской речи, обходящей Whisper в 96% сравнений: нейросеть распознавание речи

Cohere releases open-source Arabic speech recognition model outperforming Whisper

Cohere has released Cohere Transcribe Arabic, a 2-billion-parameter model specifically designed for transcribing Arabic speech. This specialized model reportedly outperforms larger, more general-purpose models like OpenAI's Whisper and Meta's OmniASR on Arabic dialects and code-switching between Arabic and English. The model is open-sourced under the Apache 2.0 license and available on Hugging Face, making it accessible for commercial use and further development. AI

IMPACT This specialized model could significantly improve the accuracy and reduce the cost of Arabic speech-to-text applications, potentially setting a new standard for niche language models.

RANK_REASON Frontier-lab model release with system card and benchmark data. [lever_c_demoted from frontier_release: ic=1 ai=1.0]

Read on dev.to — LLM tag →

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →

Cohere releases open-source Arabic speech recognition model outperforming Whisper

COVERAGE [1]

  1. dev.to — LLM tag TIER_1 Русский(RU) · Promptra Team ·

    Cohere open-sourced an Arabic speech recognition model that outperforms Whisper in 96% of comparisons: neural network speech recognition

    <p>Если ты когда-нибудь прогонял арабское аудио через Whisper и получал на выходе кашу из перепутанных диалектов и потерянных английских слов посреди фразы, у тебя теперь есть альтернатива. 7 июля 2026 года Cohere выложила модель, которая делает ровно одну вещь - переводит арабск…