Parakeet Multitalker Int8 Onnx

--- license: other license_name: nvidia-open-model-license license_link: https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/ library_name: onnx pipeline_tag: automatic-speech-recognition language: - en tags: - automatic-speech-recognition - speaker-diarization - multitalker-asr - onnx - int8 - onnxruntime - parakeet - fastconformer - recogment base_model: nvidia/multitalker-parakeet-streaming-0.6b-v1 --- # Parakeet Multitalker (int8 ONNX) — Recogment mirror Streaming joint ASR + 4-slot speaker diarization. This repo packages **int8-quantized ONNX exports** of the upstream NVIDIA NeMo checkpoint so that runtimes which don't depend on PyTorch / NeMo can use the model directly. > ⚠️ **Unofficial.** This is a community-maintained mirror of derived artefacts. For the canonical model, citation, and training details see [nvidia/multitalker-parakeet-streaming-0.6b-v1](https://huggingface.co/nvidia/multitalker-parakeet-streaming-0.6b-v1). ## Files | File | Purpose | Size | |---|---|---| | `encoder.int8.onnx` | FastConformer encoder | ~626 MB | | `decoder_joint.int8.onnx` | RNN-T decoder + joint network | ~9 MB | | `tokenizer.model` | SentencePiece tokenizer (1024 vocab) | ~245 KB | ## Provenance These ONNX files were exported from the upstream NeMo `.nemo` checkpoint of [`nvidia/multitalker-parakeet-streaming-0.6b-v1`](https://huggingface.co/nvidia/multitalker-parakeet-streaming-0.6b-v1) and then quantized to int8 for CPU inference. The conversion pipeline lives in the [parakeet-rs](https://github.com/dnth/parakeet-rs) project's tooling; the runtime that drives the resulting files is the same crate. No model weights have been retrained, fine-tuned, or otherwise modified beyond the int8 quantization step. ## Why this mirror exists The [Recogment](https://github.com/kevin-atx/recogment) daemon ships with a pinned set of model SHA-256 hashes and a zero-egress sandbox; the only outbound network in the whole product is a companion downloader binary that fetches from public HTTPS sources. Until this mirror existed, the Parakeet files had to be shipped to beta testers out-of-band as a tarball. With this mirror in place, the downloader can fetch them automatically alongside the rest of the public catalogue. If you're not building Recogment, you almost certainly want the official NVIDIA repo linked above instead — it includes t...