Multitalker Parakeet Streaming 0.6b v1 GGUF

--- license: other license_name: nvidia-open-model-license license_link: https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/ base_model: nvidia/multitalker-parakeet-streaming-0.6b-v1 base_model_relation: quantized library_name: transcribe.cpp pipeline_tag: automatic-speech-recognition language: - en tags: - gguf - transcribe.cpp - asr - speech-to-text - parakeet - conformer - rnnt - streaming - cache-aware - diarization - multitalker - sortformer transcribe_cpp: wer_librispeech_test_clean: f32: 2.19 f16: 2.19 q8_0: 2.18 q6_k: 2.2 q5_k_m: 2.18 q4_k_m: 2.18 rtf_m4_max: metal: 178.3 cpu: 34.6 rtf_ryzen_4750u: cpu: 13 vulkan: 25 cpwer_ami_ihm_test: bundle_f32_kernel: 19.35 bundle_f32_masked: 23.73 streaming: true diarize: true translate: false lang_detect: false timestamps: token --- # multitalker-parakeet-streaming-0.6b-v1: transcribe.cpp GGUF GGUF conversions of [nvidia/multitalker-parakeet-streaming-0.6b-v1](https://huggingface.co/nvidia/multitalker-parakeet-streaming-0.6b-v1) for use with [transcribe.cpp](https://github.com/handy-computer/transcribe.cpp). Ported from upstream commit [8749fc7](https://huggingface.co/nvidia/multitalker-parakeet-streaming-0.6b-v1/commit/8749fc7), pinned 2026-07-12. Validated against the NeMo reference at transcribe.cpp commit [c55a09d](https://github.com/handy-computer/transcribe.cpp/tree/c55a09d) on 2026-07-12. Offline and cache-aware streaming English speech-to-text with punctuation and capitalization. A 0.6B-parameter cache-aware streaming FastConformer encoder with an RNN-T transducer decoder, fine-tuned from nvidia/nemotron-speech-streaming-en-0.6b. Upstream this is a multitalker (speaker-attributed) checkpoint, and this repo ships that path too: the plain GGUFs run the single_speaker_mode ASR path (one flat transcript), while the **bundle GGUFs** (in `bundle/`) embed the [nvidia/diar_streaming_sortformer_4spk-v2.1](https://huggingface.co/nvidia/diar_streaming_sortformer_4spk-v2.1) streaming diarizer and, with `--diarize`, transcribe up to 4 overlapping speakers into a speaker-tagged transcript. The encoder preserves the upstream att_context_size=[70, 13] (1.12s) cache-aware attention mask; all four latency lookahead settings are selectable. ## Downloads | Quantization | Download | Size | WER (LibriSpeech test-clean, offline) | | --- | --- | ---:...

Source context: 24011 downloads · 1 likes · Pipeline automatic-speech-recognition · Library transcribe.cpp · Repo handy-computer/multitalker-parakeet-streaming-0.6b-v1-gguf