뉴스 및 전체 검색
뉴스, 도구, 모델, 가이드, 주제, 워크플로, 프롬프트를 함께 검색합니다.
AI 모델 149
“Transcript AI”
lang-id-voxlingua107-ecapa-gguf
Ported from upstream commit 0253049, pinned 2026-10-05. Validated against the SpeechBrain reference at transcribe.cpp commit aa4d0f47 on 2026-10-08. Spoken language identification over 107 languages: SpeechBrain's ECAPA-TDNN trained on VoxLingua107. NOT a transcription model: a run returns the model's language labels ranked by probability, optionally restricted to a caller-chosen set. Takes 16 kHz mono WAV; scores up to the first 30 s of a clip. Top-1 accuracy on FLEURS multilingual (3,000 utterances), scored on cpu. Measured at transcribe.cpp `cf62265a` on 2026-10-08.
Ming-3.0-Audio-Tiny-GGUF
GGUF conversion of inclusionAI/Ming-3.0-audio-tiny, pinned to revision `25b96e30be3739849bf76c0f48ac11db4a262fb3`. Upstream code: Ming-3.0-audio-flash. The single GGUF contains the audio encoder, KDA/MLA MoE Thinker, Talker, audio VAE, CAMPPlus speaker encoder, tokenizers, and configuration. It supports offline transcription, audio understanding, named voices, voice cloning, and spoken responses. Speech output is mono 44.1 kHz. Both quantized packages preserve the audio encoder, Thinker attention and MLP projections, Talker, VAE and speaker encoder. TTS-only memory usage therefore does not shrink with Thinker expert quantization.
whisper-large-v3-turbo
This model is a Hebrew finetune (continued training) of the OpenAI Whisper Large v3 Turbo model. - **Developed by:** ivrit-ai - **Language(s) (NLP):** Hebrew - **License:** Apache-2.0 - **Finetuned from model** openai/whisper-large-v3-turbo - **Training Date** Apr 2025 Language detection capability of this model has been degraded during training - it is intended for mostly-hebrew audio transcription. Language token should be explicitly set to Hebrew.
granite-speech-4.1-2b-nar-gguf
GGUF conversions of ibm-granite/granite-speech-4.1-2b-nar for use with transcribe.cpp. Ported from upstream commit 99a4df9, pinned 2026-05-24. Validated against the Transformers reference at transcribe.cpp commit c53af2c on 2026-05-24. Offline multilingual speech-to-text in a single non-autoregressive editor pass. IBM Granite Speech 4.1-2b NAR shares the Conformer audio encoder with the AR Granite-Speech family but pairs it with a custom MLP-with- attention projector and the Granite-4.0-1b LLM used as a bidirectional editor (causal mask disabled). One forward pass produces logits over the full transcript; CTC decode yields the final text. No token-by-token loop. Takes a 16 kHz mono WAV and produces a transcript. English plus French, German, Spanish, and Portuguese; ASR only (no translation, no timestamps).
needle3
A foundation model for mobiles, wearables, robots, smart home, automotive and microcontrollers. The whole model is a single 8-29 MB file, and we trade general chat capacity to beat models 10x its size on mobile tool calls and match 2-3x bigger models on extraction. - **Tool calls**: given the functions your app exposes, Needle picks the right ones and fills every argument from what the user said. Ask for two things and you get two calls in order; ask for something no tool covers and you get an empty list, not a guess. - **Structured extraction**: declare a shape, hand over messy text, get typed fields back: an invoice, a booking, a notification, a form. The decode grammar guarantees the output parses, and extraction generalises to classification. - **Text embedding**: the same model returns a vector for a sentence, so an app can search, match and route locally. The same engine also runs Whistle, our 16.9 MB speech-to-text model in seven languages. Load `whistle.cact` beside `needle3.cact` and the one binary transcribes a clip on its own, or answers it against your tools in a single call and returns the calls and the transcript in one JSON object.
kosmos-2.5
Kosmos-2.5 is a multimodal literate model for machine reading of text-intensive images. Pre-trained on large-scale text-intensive images, Kosmos-2.5 excels in two distinct yet cooperative transcription tasks: (1) generating spatially-aware text blocks, where each block of text is assigned its spatial coordinates within the image, and (2) producing structured text output that captures styles and structures into the markdown format. This unified multimodal literate capability is achieved through a shared decoder-only auto-regressive Transformer architecture, task-specific prompts, and flexible text representations. We evaluate Kosmos-2.5 on end-to-end document-level text recognition and image-to-markdown text generation. Furthermore, the model can be readily adapted for any text-intensive image understanding task with different prompts through supervised fine-tuning, making it a general-purpose tool for real-world applications involving text-rich images. This work also paves the way for the future scaling of multimodal large language models. Since this is a generative model, there is a risk of **hallucination** during the generation process, and it **CAN NOT** guarantee the accuracy of all OCR/Markdown results in the images.
whisper-large-v3-turbo-german
This model map provides information about a model based on Whisper Large v3 that has been fine-tuned for speech recognition in German. Whisper is a powerful speech recognition platform developed by OpenAI. This model has been specially optimized for processing and recognizing German speech. - Transcription of spoken German language - Voice commands and voice control - Automatic subtitling for German videos - Voice-based search queries in German - Dictation functions in word processing programs The data and code for evaluations are available here
sage-fredt5-distilled-95m
The model corrects spelling and punctuation errors and typos by bringing all the words in the text to the norm of the Russian language. Corrector is a distilled version of the original model that had been trained based on the FRED-T5-1.7B architecture. An extensive dataset with “artificial” errors was taken as a training corpus: the corpus was assembled on the basis of the Russian-language Wikipedia and transcripts of Russian-language videos, then typos and spelling errors were automatically introduced into it using the library SAGE. Below are automatic metrics for determining the correctness of the spell checkers. We compare our solution with both open automatic spell checkers and the ChatGPT family of models on all four available datasets: - **RUSpellRU**: texts collected from (LiveJournal), with manually corrected typos and errors; - **MultidomainGold**: examples from 7 text sources, including the open web, news, social media, reviews, subtitles, policy documents and literary works; - **MedSpellChecker**: texts with errors from medical anamnesis; - **GitHubTypoCorpusRu**: spelling errors and typos in commits from GitHub; **RUSpellRU** | Model | Pr. (spell) | Rec. (spell) | F1 (spell) | Pr. (punc) | Rec. (punc) | F1 (punc) | Pr. (case) | Rec. (case) | F1 (case) | | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | | sage-fredt5-distilled-95m | 83.5 | 74.8 | 78.9 | 86.8 | 80.6 | 83.6 | 94.4 | 92.5 | 93.5 | | sage-ai-service | 90.3 | 86.3 | 88.2 | 90.3 | 86.6 | 88.4 | 95.2 | 95.9 | 95.6 | | gpt-3.5-turbo | 33.6 | 58.5 | 42.7 | 85.9 | 64.6 | 73.7 | 84.9 | 73.9 | 79.0 | | gpt-4 | 54.9 | 76.7 | 64.0 | 84.0 | 82.3 | 83.2 | 91.5 | 90.2 | 90.9 |
whistle
A speech-to-text model for mobiles, wearables, robots, smart home, automotive and microcontrollers. The whole model is a single 16.9 MB file, and it runs on the same CPU engine as Needle, from the same container and the same quantisation, with no dependencies and no GPU. - **Transcription**: 16 kHz mono audio of any length, one pass up to 30 seconds and windowed beyond it, in English, German, French, Spanish, Italian, Dutch and Polish. The language is detected unless you name it, and silence and steady noise return an empty transcript rather than an invented sentence; other sounds, music especially, can still come out as text, so put a voice activity detector or a wake word in front of an always-on microphone. Live audio streams for as long as you like, committing the words two passes agree on. - **Word timestamps**: every word with its start, end and probability, aligned from the decoder's own attention, so an app can highlight, seek or cut on a word. - **Speech embedding**: the encoder output, one row per 80 ms frame, for matching and retrieval without decoding a transcript at all. Keyword biasing takes the names, places, product words and jargon your users actually say and favours them during the search, so the ones that matter survive.