News and site search
Explore related news, tools, models, guides, topics, workflows and prompts.
AI Models 149
“Transcript AI”
Fun-ASR-Nano-2512-gguf
GGUF conversions of FunAudioLLM/Fun-ASR-Nano-2512 for use with transcribe.cpp. Ported from upstream commit a7088d6, pinned 2026-05-06. Validated against the FunASR reference at transcribe.cpp commit f094d28 on 2026-05-06. Offline speech-to-text in Chinese, English, and Japanese, plus 7 Chinese dialects (Wu, Cantonese, Min, Hakka, Gan, Xiang, Jin) and 26 regional Mandarin accents. ~800M trainable parameters wrapping a frozen SenseVoiceEncoderSmall (50 SAN-M main blocks + 20 transformer blocks), a 2-layer audio adaptor (512 → 1024), and a bundled Qwen3-0.6B LLM (28 layers, 16/8 GQA, BF16) that produces the transcript autoregressively. Takes a 16 kHz mono WAV and emits text. Not a streaming model, no translation, no built-in long-form chunking, no timestamps. ITN (inverse text normalization) is off by default, since with it on English can come back all caps; pass `--itn` for digits and punctuation, which helps Chinese.
kotoba-whisper-bilingual-v1.0
_Kotoba-Whisper-Bilingual_ is a collection of distilled Whisper models trained for - **Japanese ASR** - **English ASR** - **Speech-to-text translation (Japanese -> English)** - **Speech-to-text translation (English -> Japanese)** developed through the collaboration bewteen Asahi Ushio and Kotoba Technologies. Following the original work of distil-whisper (Robust Knowledge Distillation via Large-Scale Pseudo Labelling), we employ OpenAI's Whisper large-v3 as the teacher model for Japanese and English ASR, while we translate the transcription into English and Japanese by external LLM to obtain training dataset for speech-to-text translation. We employ ReazonSpeech for Japanese ASR and Japanese speech to English text translation, and Multilingual LibriSpeech for English ASR and English speech to Japanese text translation. Kotoba-whisper-bilingual's loss objective consists of cross-entropy on both of ASR and translation tasks, while KL divergence loss only for ASR task. The student model consists the full encoder of the teacher large-v3 model and the decoder with two layers initialized from the first and last layer of the large-v3 model. As kotoba-whisper uses the same architecture as distil-whisper/distil-large-v3, it inherits the benefit of the improved latency compared to openai/whisper-large-v3 (**6.3x faster than large-v3**, see the table below taken from distil-whisper/distil-large-v3).
qwen3-forced-aligner-0.6b
**Word-level forced alignment for OpenASR transcripts -- a non-autoregressive Qwen3 audio+text model that refines per-word timestamps** A capability-pack support model for the **OpenASR** runtime — pure-Rust inference, **no Python at inference time**. Not a standalone transcription model: it augments another OpenASR ASR model's own decode path. Qwen3-ForcedAligner-0.6B is a **word-level forced-alignment** model from **Qwen**, sharing its audio-encoder + LM `thinker` tensor layout byte-for-byte with **Qwen3-ASR** (same `Qwen3ASRForConditionalGeneration` architecture). The only structural difference is the final head: instead of a tied vocabulary `lm_head`, it uses an independent `Linear(hidden_size, 5000)` classification head over 80ms-wide timestamp bins. Given a transcript's text and its source audio, it runs a single non-autoregressive forward pass and reads off word-boundary timestamps at argmax ` ` positions -- refining a model family's own (typically decode-time-approximate) per-word timestamps. This OpenASR repo repackages the weights as `.oasr` packs that run natively in the OpenASR runtime -- no Python at inference, all decoding local. OpenASR recommends the validated **q4_k** tier and also ships **q8_0** and an **fp16** full-precision reference tier. The q4_k label is the catalog's unified product name, not a claim that every matrix uses Q4_K: the audio encoder, token embedding, and timestamp head remain Q8_0, and pack verification replays the exact per-tensor policy. Q3 and legacy all-Q4 packs are rejected because small logit perturbations can move a word boundary across multiple 80ms bins.
Qwen3-ASR-0.6B-hf
The Qwen3-ASR family includes **Qwen3-ASR-1.7B** and **Qwen3-ASR-0.6B**, which support language identification and ASR for 52 languages and dialects. Both leverage large-scale speech training data and the strong audio understanding capability of their foundation model, Qwen3-Omni. The 1.7B version achieves state-of-the-art performance among open-source ASR models and is competitive with the strongest proprietary commercial APIs. - **All-in-one:** Supports language identification and speech recognition for 30 languages and 22 Chinese dialects, including English accents from multiple countries and regions. - **Excellent and Fast:** High-quality and robust recognition under complex acoustic environments. Qwen3-ASR-0.6B reaches 2000× throughput at a concurrency of 128. Both models support streaming/offline unified inference with a single model and handle long audio. - **Forced Alignment:** Qwen3-ForcedAligner-0.6B supports timestamp prediction for arbitrary units within up to 5 minutes of speech in 11 languages, surpassing E2E-based forced-alignment models in accuracy. `apply_transcription_request` handles chat-template formatting for you and is the recommended entry point.
muscriptor-medium
**MuScriptor** is an open-weight model for **general-purpose, multi-instrument automatic music transcription (AMT)**: it converts a music recording (any genre, multiple simultaneous instruments) into a stream of notes played. This repository hosts the **medium** variant (≈300M parameters), the default checkpoint downloaded by the `muscriptor` library. `muscriptor-medium` balances quality and footprint. For the best transcription quality use `muscriptor-large` (≈1.3B); for the smallest/fastest option use `muscriptor-small` (≈100M). `load_model` accepts a size keyword (`"small"`/`"medium"`/`"large"`), a local `.safetensors` path, or an `hf://` / `https://` URL. Weights loaded by size keyword (or any `hf://` URL) are cached in the standard Hugging Face cache (`~/.cache/huggingface/hub`, configurable via `HF_HOME`); weights fetched from a plain `http(s)://` URL are cached under `~/.cache/muscriptor/`. Input audio can be WAV or any format `libsndfile` reads (mp3, flac, ogg, m4a, …); it is resampled to 16 kHz mono internally.
wav2vec2-french-phonemizer-v2
This version fixes the absence of the voiced labial-palatal approximant **ɥ**. The original Cnam-LMSSC/wav2vec2-french-phonemizer never transcribes the pronunciation of the voiced labial–palatal approximant with /ɥ/ and instead incorrectly replaces it with the sequence /yi/. This created issues for phonetic accuracy and linguistic analysis. Words like "nuit" or "huit" are pronounced with \ɥ\ and not with \yi\ For Text-to-Speech (TTS) task, substituting \ɥ\ with \yi\ results in unnatural prosody. Fine-tuned facebook/wav2vec2-base-fr-voxpopuli-v2 for **French speech-to-phoneme** (without language model) using the train and validation splits of Common Voice v13. As this model is specifically trained for a speech-to-phoneme task, the output is sequence of IPA-encoded words, without punctuation. If you don't read the phonetic alphabet fluently, you can use this excellent IPA reader website to convert the transcript back to audio synthetic speech in order to check the quality of the phonetic transcription.
VibeVoice-ASR
**VibeVoice-ASR** is a unified speech-to-text model designed to handle **60-minute long-form audio** in a single pass, generating structured transcriptions containing **Who (Speaker), When (Timestamps), and What (Content)**, with support for **Customized Hotwords** and over **50 languages**. ➡️ **Code:** microsoft/VibeVoice ➡️ **Demo:** VibeVoice-ASR-Demo ➡️ **Report:** VibeVoice-ASR Technical Report ➡️ **Finetuning:** Finetuning ➡️ **vLLM:** vLLM-VibeVoice-ASR - **🕒 60-minute Single-Pass Processing**: Unlike conventional ASR models that slice audio into short chunks (often losing global context), VibeVoice ASR accepts up to **60 minutes** of continuous audio input within 64K token length. This ensures consistent speaker tracking and semantic coherence across the entire hour.
Nemotron-3-Diarization-GGUF
**Update 2026-09-26**: Since the initial release, Nemotron 3 Diarization has received improved NeMo parity for ragged batches and streaming tails, selectable flash/eager attention, self-describing speaker-probability output, ASR-compatible latency profiles, and integration with Nemotron’s speaker-tagged transcription pipeline. Existing published GGUFs remain compatible. GGUF weights for native inference with audio.cpp. Tested on CUDA, Vulkan, and Metal(Mac) backends. Nemotron 3 Diarization identifies who spoke when, with up to eight speakers. It supports offline, streaming, and server batch diarization. It returns speaker turns and timestamps, not transcribed words. This package was converted from the original `.nemo` checkpoint, not from an existing GGUF. Each file contains 362 tensors, frontend configuration, and an embedded model spec, with no separate weight or configuration files required. The BF16 file retains the original model weight precision, with the frontend mel filter stored in F32. BF16 is parity safe with Nemo with TF32 control.
granite-speech-4.1-2b-plus-gguf
Ported from upstream commit edd3bf5, pinned 2026-05-17. Validated against the Transformers reference at transcribe.cpp commit 275332d on 2026-05-17. Offline multilingual speech-to-text with word-level timestamps. IBM Granite Speech 4.1-2b-plus is the timestamp-and-diarization variant of the Granite-Speech family. Same architecture as the base 4.1-2b (Conformer encoder, BLIP-2 Q-Former projector, Granite-4.0-1b autoregressive LLM decoder) with two changes: the encoder concatenates mid-layer (idx 3) and final-layer hidden states (`cat_hidden_layers=[3]`, doubling the projector K/V input from 1024 to 2048), and the LM token embeddings are tied with the lm_head. Takes a 16 kHz mono WAV and produces a transcript, with `--timestamps word` returning structured per-word timestamps (parsed from the model's `[T:N]` centisecond markers), or `--diarize` returning structured speaker-attributed turns from its separate SAA prompt. Those two prompt tasks cannot be combined. Transcribes English, French, German, Spanish, and Portuguese (no Japanese on this variant). This variant is transcription-only: unlike the base granite-speech-4.1-2b, it does not perform speech translation. WER on the full LibriSpeech test-clean split (2,620 utterances), batch size 1, timestamps none. Figures without a commit were published before provenance was recorded.
llmlingua-2-bert-base-multilingual-cased-meetingbank
This model was introduced in the paper **LLMLingua-2: Data Distillation for Efficient and Faithful Task-Agnostic Prompt Compression** (Pan et al, 2024). It is a BERT multilingual base model (cased) finetuned to perform token classification for task agnostic prompt compression. The probability `$p_{preserve}$` of each token `$x_i$` is used as the metric for compression. This model is trained on the extractive text compression dataset constructed with the methodology proposed in the **LLMLingua-2**, using training examples from MeetingBank (Hu et al, 2023) as the seed data. You can evaluate the model on downstream tasks such as question answering (QA) and summarization over compressed meeting transcripts using this dataset. For more details, please check the project page of LLMLingua-2 and LLMLingua Series.