Hanzo speech
Copyright 2023-2026 Hanzo AI, Inc.

This service is licensed under the Apache License, Version 2.0 (LICENSE).
It serves the models and links the software below. Weights are not in this
repository or in the image; weights.sum names every file, fetched at boot.

Models
------

parakeet — NVIDIA Parakeet TDT 0.6B v3
  Copyright NVIDIA Corporation. Licensed under CC BY 4.0
  (https://creativecommons.org/licenses/by/4.0/).
  Source: https://huggingface.co/nvidia/parakeet-tdt-0.6b-v3
  Changes: converted to ONNX and quantized to int8 by the k2-fsa sherpa-onnx
  project (sherpa-onnx-nemo-parakeet-tdt-0.6b-v3-int8,
  https://github.com/k2-fsa/sherpa-onnx/releases/tag/asr-models); served
  unmodified from that export.

whisper — OpenAI Whisper large-v3-turbo
  Copyright (c) 2022 OpenAI. Licensed under the MIT License.
  Source: https://github.com/openai/whisper
  Changes: encoder from the k2-fsa sherpa-onnx int8 export
  (sherpa-onnx-whisper-turbo), with the alignment-head metadata added; decoder
  re-exported by Hanzo with sherpa-onnx's export-onnx-with-attention.py, which
  adds the alignment heads' cross-attention as an output, and quantized to int8.

Language identification — OpenAI Whisper tiny
  Copyright (c) 2022 OpenAI. Licensed under the MIT License.
  Source: https://github.com/openai/whisper, int8 export
  sherpa-onnx-whisper-tiny by k2-fsa.

kokoro — Kokoro-82M v1.0, 54 voices
  Copyright hexgrad. Licensed under the Apache License, Version 2.0.
  Source: https://huggingface.co/hexgrad/Kokoro-82M; ONNX export
  kokoro-multi-lang-v1_0 by k2-fsa sherpa-onnx.

Voice activity — Silero VAD
  Copyright (c) 2020-present Silero Team. Licensed under the MIT License.
  Source: https://github.com/snakers4/silero-vad

Software
--------

sherpa-onnx — Copyright k2-fsa. Apache License 2.0.
  https://github.com/k2-fsa/sherpa-onnx (Go bindings sherpa-onnx-go-linux)
ONNX Runtime — Copyright Microsoft Corporation. MIT License.
  https://github.com/microsoft/onnxruntime
espeak-ng — GPL-3.0-or-later. Built into sherpa-onnx's text-to-speech
  front end, with its data (espeak-ng-data) shipped beside the Kokoro weights.
  Source: https://github.com/espeak-ng/espeak-ng
FFmpeg — LGPL-2.1-or-later / GPL-2.0-or-later as built by Debian.
  https://ffmpeg.org, invoked as a separate program.
