Cohere zaprezentowało Transcribe – model ASR z 2 miliardami parametrów, który detronizuje Whisper Large V3 w szybkości i precyzji, radząc sobie z wyzwaniami dialektów arabskich.

Description: Snap Video Translator brings every step of localizing a video into a single Windows app. Load a video and it takes you straight through transcription, AI translation, and either subtitle burn-in or an AI voice-over. Transcription (runs locally)Runs OpenAI Whisper on your own PC. Choose from small / medium / large-v3-turbo / large-v3 to balance speed and accuracy. Your audio is never sent to an external server for transcription. AI translationWorks with AI services such as Gemini, OpenAI, and Claude, or with an OpenAI-compatible local LLM. Produces natural translations of your video's content. Subtitle burn-inAdjust font size, text color, outline, and position, then burn subtitles directly into the video. AI dubbingConverts the translation into speech and generates a voice-over in 7 languages (Japanese, English, Chinese, Korean, French, German, Spanish). Mix it with the original audio or replace it. Subtitle file exportExport subtitle files in SRT / VTT format. Review & edit modeCheck the subtitle text on screen before burn-in and edit it if needed. Batch processingProcess multiple videos at once. Who it's for- Video creators who want to reach viewers in other languages- Education and training teams localizing lectures and tutorials- Companies localizing internal videos Before you start- Transcription runs locally with Whisper; your audio is not sent anywhere.- Using a cloud AI service for translation requires that service's API key. You can also use an OpenAI-compatible local LLM.- AI translation and AI dubbing require an internet connection (no API key is needed for dubbing).- Subtitles and dubbed audio are AI-generated. Please review the output before public or business use.- FFmpeg is bundled as the video processing engine. System RequirementsWindows 10 version 19041.0 or higher (64-bit) * Japanese / English UI, selected automatically. Release Name: Snap Video Translator 1.1.1Size: 395.6 MBLinks: HOMEPAGE – NFO – Torrent Search Download: RAPiDGATOR

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

<p>Please, help!</p> <p>Although the default Whisper API setting is <strong>Transcribing</strong>, I receive a <strong>Translation</strong> (besides into different random languages!?!). Both whisper-large-v3 and whisper-large-v3-turbo. Why?</p> <p>final request = http.Request(‘POST’, url); <a href="//router.huggingface.co/hf-inference/models/openai/whisper-large-v3-turbo">//router.huggingface.co/hf-inference/models/openai/whisper-large-v3-turbo</a><br /> request.headers[‘Authorization’] = ‘Bearer $_hfToken’;<br /> request.headers[‘Content-Type’] = ‘audio/m4a’;<br /> request.bodyBytes = audioBytes;</p> <p>*** I tried to add additional headers:</p> <p>request.headers[‘Accept-Language’] = ‘en,en-US’;<br /> request.headers[‘language’] = ‘en’;<br /> request.headers[‘task’] = ‘transcribe’;</p> <p>… but it didn’t help.</p> <p>*** I also tried through json payload like:<br /> {<br /> ‘inputs’: base64Audio,<br /> ‘parameters’: {<br /> ‘task’: ‘transcribe’,<br /> ‘language’: ‘en’,<br /> },<br /> }</p> <p>… but returned an error "unexpected keyword argument ‘task’ "… yes ‘task’ is not in the public HF api’s parameters but it’s strange that there are no standard ones of the Whisper API.</p> <p>How can I fix this issue? I need to receive always and only Transcribe (not Translate to random language) as stated as the default behavior in Whisper Hugging Face Inference API.</p>

Accurate documentation is critical in ophthalmology, yet clinical notes often contain subtle errors that can affect decision-making. This study prospectively compared contemporary large language models (LLMs) for detecting clinically salient errors in emergency ophthalmology encounter notes and generating actionable corrections. 129 de-identified notes, each seeded with a predefined target error, were independently audited by four LLMs (o3 (OpenAI, closed-source), DeepSeek-v3-r1 (Deepseek, open-source), MedGemma-27B (Google, open-source), and GPT-4o (OpenAI, closed-source)) using a standardized prompt. Two masked ophthalmologists graded error localization, relevance of additional issues, and overall recommendation quality, with within-case analyses applying appropriate nonparametric tests. Performance varied significantly across models (Cochran&rsquo;s Q = 71.13, p = 2.44 &times; 10&minus;15). o3 achieved the highest error localization accuracy at 95.7% (95% CI, 89.5&ndash;98.8), followed by DeepSeek-v3-r1 (90.3%), MedGemma-27b (80.9%), and GPT-4o (53.2%). Ordinal outcomes similarly favored o3 and DeepSeek-v3-r1 (both p &lt; 10&minus;9 vs. GPT-4o), with mean recommendation quality scores of 3.35, 3.05, 2.54, and 2.11, respectively. These findings demonstrate that LLMs can serve as accurate &ldquo;second-eyes&rdquo; for ophthalmology documentation. A proprietary model led on all metrics, while a strong open-source alternative approached its performance, offering potential for privacy-preserving on-premise deployment. Clinical translation will require oversight, workflow integration, and careful attention to ethical considerations.

Background: Deepseek-R1, an open-source large language model (LLM), has generated significant global interest in the past months. Objective: This study aimed to compare the performance of DeepSeek and OpenAI LLMs on the Chinese National Medical Licensing Examination (NMLE) and evaluate their potential in medical education #mededu. Methods: This cross-sectional study assessed 2 DeepSeek models (DeepSeek-R1 and DeepSeek-V3), 3 OpenAI models (ChatGPT-o1 pro, ChatGPT-o3 mini, and GPT-4o), and 2 additional Chinese LLMs (ERNIE 4.5 Turbo and Qwen 3) using the 2021 NMLE. Model performance was evaluated based on overall accuracy, accuracy across question types (A1, A2, A3 and A4, and B1), case analysis and non–case analysis questions, medical specialties, and accuracy consensus between different model combinations. Results: All LLMs successfully passed the NMLE. DeepSeek-R1 achieved the highest accuracy (573/597, 96%), followed by DeepSeek-V3 (558/600, 93%), both of which significantly outperformed ChatGPT-o1 pro (450/600, 75%), ChatGPT-o3 mini (455/600, 75.8%), and GPT-4o (452/600, 75.3%; P

VOICE → TEXT

> Attempted request to https://router.huggingface.co/hf-inference/models/openai/whisper-large-v3-turbo failed: HTTP/1.1 400 Bad Request - {“error”:"\n Content type “application/octet-stream” not supported.\n Supported content types are:\n application/json, application/json; charset=UTF-8, text/csv, text/plain, image/png, image/jpeg, image/jpg, image/tiff, image/bmp, image/gif, image/webp, image/x-image, audio/x-flac, audio/flac, audio/mpeg, audio/x-mpeg-3, audio/wave, audio/wav, audio/x-wav, audio/ogg, audio/x-audio, audio/webm, audio/webm;codecs=opus, audio/AMR, audio/amr, audio/AMR-WB, audio/AMR-WB+, audio/m4a, audio/x-m4a\n "} Seems like the unity plugin only supports sending bytes while the model itself doesn’t accepts it

> Attempted request to https://router.huggingface.co/hf-inference/models/openai/whisper-large-v3-turbo failed: HTTP/1.1 400 Bad Request - {“error”:"\n Content type “application/octet-stream” not supported.\n Supported content types are:\n application/json, application/json; charset=UTF-8, text/csv, text/plain, image/png, image/jpeg, image/jpg, image/tiff, image/bmp, image/gif, image/webp, image/x-image, audio/x-flac, audio/flac, audio/mpeg, audio/x-mpeg-3, audio/wave, audio/wav, audio/x-wav, audio/ogg, audio/x-audio, audio/webm, audio/webm;codecs=opus, audio/AMR, audio/amr, audio/AMR-WB, audio/AMR-WB+, audio/m4a, audio/x-m4a\n "} Seems like the unity plugin only supports sending bytes while the model itself doesn’t accepts it

> Attempted request to https://router.huggingface.co/hf-inference/models/openai/whisper-large-v3-turbo failed: HTTP/1.1 400 Bad Request - {“error”:"\n Content type “application/octet-stream” not supported.\n Supported content types are:\n application/json, application/json; charset=UTF-8, text/csv, text/plain, image/png, image/jpeg, image/jpg, image/tiff, image/bmp, image/gif, image/webp, image/x-image, audio/x-flac, audio/flac, audio/mpeg, audio/x-mpeg-3, audio/wave, audio/wav, audio/x-wav, audio/ogg, audio/x-audio, audio/webm, audio/webm;codecs=opus, audio/AMR, audio/amr, audio/AMR-WB, audio/AMR-WB+, audio/m4a, audio/x-m4a\n "} Seems like the unity plugin only supports sending bytes while the model itself doesn’t accepts it

> Attempted request to https://router.huggingface.co/hf-inference/models/openai/whisper-large-v3-turbo failed: HTTP/1.1 400 Bad Request - {“error”:"\n Content type “application/octet-stream” not supported.\n Supported content types are:\n application/json, application/json; charset=UTF-8, text/csv, text/plain, image/png, image/jpeg, image/jpg, image/tiff, image/bmp, image/gif, image/webp, image/x-image, audio/x-flac, audio/flac, audio/mpeg, audio/x-mpeg-3, audio/wave, audio/wav, audio/x-wav, audio/ogg, audio/x-audio, audio/webm, audio/webm;codecs=opus, audio/AMR, audio/amr, audio/AMR-WB, audio/AMR-WB+, audio/m4a, audio/x-m4a\n "} Seems like the unity plugin only supports sending bytes while the model itself doesn’t accepts it

> Attempted request to https://router.huggingface.co/hf-inference/models/openai/whisper-large-v3-turbo failed: HTTP/1.1 400 Bad Request - {“error”:"\n Content type “application/octet-stream” not supported.\n Supported content types are:\n application/json, application/json; charset=UTF-8, text/csv, text/plain, image/png, image/jpeg, image/jpg, image/tiff, image/bmp, image/gif, image/webp, image/x-image, audio/x-flac, audio/flac, audio/mpeg, audio/x-mpeg-3, audio/wave, audio/wav, audio/x-wav, audio/ogg, audio/x-audio, audio/webm, audio/webm;codecs=opus, audio/AMR, audio/amr, audio/AMR-WB, audio/AMR-WB+, audio/m4a, audio/x-m4a\n "} Seems like the unity plugin only supports sending bytes while the model itself doesn’t accepts it

> Attempted request to https://router.huggingface.co/hf-inference/models/openai/whisper-large-v3-turbo failed: HTTP/1.1 400 Bad Request - {“error”:"\n Content type “application/octet-stream” not supported.\n Supported content types are:\n application/json, application/json; charset=UTF-8, text/csv, text/plain, image/png, image/jpeg, image/jpg, image/tiff, image/bmp, image/gif, image/webp, image/x-image, audio/x-flac, audio/flac, audio/mpeg, audio/x-mpeg-3, audio/wave, audio/wav, audio/x-wav, audio/ogg, audio/x-audio, audio/webm, audio/webm;codecs=opus, audio/AMR, audio/amr, audio/AMR-WB, audio/AMR-WB+, audio/m4a, audio/x-m4a\n "} Seems like the unity plugin only supports sending bytes while the model itself doesn’t accepts it

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

要約 日本語オンリーならruri v3 (わずか37mでOpenAIのtext-embedding-large-v3超え) もしかしたら日英だったらベターかも 多言語+コードならgranite-embedding はじめに LLMの普及からはや数年、175Bとかいう途方もないパラメータで動いていたLLMもいつの間にか4bに収まるようになり、スマホやPCで簡単に動かせるようになりました（現在だとQwen3-4b-thinking-2507などはかなり高性能です）。 一方、embeddingモデルはといえば、OpenAIはtext-embeddings-3-small/la... Source link

要約日本語オンリーならruri v3 (わずか37mでOpenAIのtext-embedding-large-v3超え)もしかしたら日英だったらベターかも多言語+コードならgranite-embedding はじめにLLMの普及からはや数年、175Bとかいう途方もないパラメータで動いていたLLMもいつの間にか4bに収まるようになり、スマホやPCで簡単に動かせるようになりました（現在だとQwen3