全站搜索
AI 模型 149
- Audar-ASR-V1-Turbo↗
audarai
🧭 Overview · 📊 Benchmarks · 💻 GGUF Deploy · 🤗 Transformers · 🎙️ Streaming · ⚡ vLLM · 📄 Tech Report · 🐙 GitHub · ☁️ Audar API · 📜 License **Audar-ASR-V1-Turbo** is an **Arabic-first generative speech-recognition model** — the accuracy tier of the Audar-ASR family. It recasts transcription as **audio-conditioned next-token prediction** over a unified text vocabulary (a language-model decoder rather than a CTC or transducer objective), and is built on a permissively-licensed open-weight audio-LLM foundation and adapted in-house — the contribution is the adaptation (the data curriculum and the alignment rubric), not the foundation: The result is **state-of-the-art dialectal Arabic ASR** — the lowest average WER *and* CER of any evaluated system on the *Open Universal Arabic ASR Leaderboard*. It transcribes MSA and every major Arabic dialect, code-switched Arabic–English, and English, across **30 languages** in total.
- granite-speech-4.1-2b-gguf↗
handy-computer
GGUF conversions of ibm-granite/granite-speech-4.1-2b for use with transcribe.cpp. Ported from upstream commit 8f4bb5f, pinned 2026-05-17. Validated against the Transformers reference at transcribe.cpp commit 275332d on 2026-05-17. Offline multilingual speech-to-text. IBM Granite Speech 4.1-2b is an audio-LLM with the same architecture as 4.0-1b (Conformer encoder with block-local Shaw attention, BLIP-2 Q-Former projector, Granite-4.0-1b-base autoregressive LLM decoder) and improved punctuation and casing over 4.0-1b. Takes a 16 kHz mono WAV and produces a transcript. Transcribes English, French, German, Spanish, Portuguese, and Japanese. Translates between English and each of those five other languages in either direction (en ↔ fr, en ↔ de, en ↔ es, en ↔ pt, en ↔ ja) — always via English, no direct fr↔de etc.
- Qwen3.8-Flash-Next-MLX-Serve-iQ-MLX-4.7bpw↗
ddalcu
mlx-serve pack of Qwen/Qwen3.8-Flash-Next, the Qwen4 preview architecture (`model_type: qwen4_exp`). Runs on a 128 GB Mac with about 75 GB resident. Includes the MTP head and the vision tower (image and video input). This is the calibrated twin of Qwen3.8-Flash-Next-MLX-Serve-mixed-4-8bit: the same layout and the same size, byte for byte (every expert 4-bit group 64, the rest 8-bit), but every expert is quantized with an importance matrix collected on the bf16 model, so the scales follow the activations the weights actually see. Same speed, same memory, closer to bf16. The importance matrix was collected on the bf16 checkpoint over 1.35 M tokens of agent transcripts, code (including SWE-bench Lite issues and patches), prose and math, every expert of every layer hit. Quality is scored against the bf16 model's own logits on 290 held-out positions (top-1: the pack's greedy token equals bf16's; KLD: KL(bf16 || pack) over bf16's top-1024 tokens; the KLD mean is dominated by a few flat prose positions, so the median is the steadier number):
- LightOnOCR-3-4B↗
lightonai
**Best OCR model.** LightOnOCR-3-4B is the largest and most accurate model of the LightOnOCR-3 family, **LightOn's** new generation of lightweight end-to-end OCR models with layout and visual understanding. We recommend it for most OCR tasks. LightOnOCR-3 is a new family of highly performant lightweight OCR models. Compared to the previous generation, they bring significant improvements in speed and transcription quality and introduce new visual understanding features: the models now output bounding box coordinates with labels for all visual elements of a document, short descriptions of images, and the numerical data of figures and charts. With these capabilities, they offer a ready-to-use, easier-to-maintain alternative to complex document understanding pipelines. The models come in three sizes. The 1B keeps the LightOnOCR-2-1B architecture, while the 0.8B and 4B adopt the Qwen3.5 vision-language architecture, which simplifies integration with existing tools and brings a significant speed-up. All models are released under the Apache 2.0 license for research and commercial use.
- Dia-1.6B↗
nari-labs
Dia is a 1.6B parameter text to speech model created by Nari Labs. It was pushed to the Hub using the PytorchModelHubMixin integration. Dia **directly generates highly realistic dialogue from a transcript**. You can condition the output on audio, enabling emotion and tone control. The model can also produce nonverbal communications like laughter, coughing, clearing throat, etc. To accelerate research, we are providing access to pretrained model checkpoints and inference code. The model weights are hosted on Hugging Face. The model only supports English generation at the moment.
- canary-ctc-aligner-GGUF↗
cstr
GGUF / ggml conversions of the **auxiliary CTC alignment model** that ships *inside* `nvidia/canary-1b-v2`'s `.nemo` tarball, repackaged as a standalone forced-aligner for use with the `nfa-align` CLI from **CrispStrobe/CrispASR@parakeet**. When you download `nvidia/canary-1b-v2` you get a single `canary-1b-v2.nemo` tarball. Inside that tarball there are TWO weight files: That second model is what NVIDIA calls the "auxiliary CTC model" used by NeMo Forced Aligner (NFA) to compute frame-aligned word and segment timestamps for Canary's transcript output. **It's the official NVIDIA-recommended path for getting word-level timestamps from Canary**, since the main encoder–decoder doesn't emit them natively (cross-attention DTW tops out around ~360 ms MAE; CTC forced alignment gives ~30-80 ms).
- LFM2-2.6B-Transcript↗
LiquidAI
- canary-1b-v2-gguf↗
handy-computer
GGUF conversions of nvidia/canary-1b-v2 for use with transcribe.cpp. Ported from upstream commit 87bc526, pinned 2026-05-08. Validated against the NeMo reference at transcribe.cpp commit db53eda on 2026-05-08. Offline multilingual speech-to-text and translation across 25 European languages. A multitask AED with a 32-layer FastConformer encoder and an 8-layer Transformer decoder. Supports automatic speech recognition for any of the 25 supported languages, plus translation between supported language pairs (per the upstream model card). Takes a 16 kHz mono WAV and produces a transcript. Not a streaming model; word and segment timestamps from the upstream model are not exposed in the v1 port.
- qwen3-forced-aligner-0.6b-GGUF↗
cstr
GGUF conversions of `Qwen/Qwen3-ForcedAligner-0.6B` — a single-pass forced aligner that takes any (audio, transcript) pair and predicts per-word/per-token timestamps. Reuses the Qwen3-ASR audio encoder + 28-layer LLM body but swaps the lm_head from `(vocab, d)` to `(5000, d)`: each ` ` placeholder you embed in the input gets a 5000-class softmax over `class * 80 ms` timestamps. Plug it into CrispASR via `-am qwen3-forced-aligner-*.gguf` to get word-level timing on **any** transcription backend — voxtral, voxtral4b, qwen3-asr, granite, parakeet, canary, cohere, even whisper. It's an alternative to the existing `canary-ctc-aligner` second-pass with broader language coverage and 80-ms resolution. The CrispASR C++ runtime auto-detects which variant a loaded GGUF is by reading the lm_head shape from `output.weight.ne[1]` — no separate backend, no separate library.
- typhoon-asr-qwen-0.6b-ctx↗
typhoon-ai
Thai ASR with **prompt-based contextual biasing**: a LoRA fine-tune (merged, 0.80B params) of `Qwen/Qwen3-ASR-0.6B-hf` that accepts a list of names/terms in the system prompt, boosts the ones that are actually spoken, writes them with the **listed spelling**, and ignores the rest (distractor false alarm rate ≤.0004; non-speech probe 5/5). The economical tier: 2.6× smaller than the 1.7b for ~.09 recall. Plain transcription (CER%, same protocol: gigaspeech2 bench in-domain / TVSpeech out-of-domain 30-s talk-show): † single-digit repetition-loop utterances dominate (0.6b: one utt; Thonburian: three, excluding them 8.39).