Open-source acknowledgements
Saidwire is built on open-source models and software. We're grateful to the people who made them. This page gives each one's licence and the attribution its licence asks for. Last updated: October 28, 2026.
Models
| Model | Licence | Attribution and changes |
|---|---|---|
| Parakeet-TDT 0.6B v2 NVIDIA |
CC BY 4.0 | "parakeet-tdt-0.6b-v2" by NVIDIA, licensed under CC BY 4.0. Changes: none to the model weights. We run the model in our own streaming inference code (16-bit precision, audio processed in short overlapping windows) and post-process its output. |
| Parakeet-CTC 0.6B NVIDIA |
CC BY 4.0 | "parakeet-ctc-0.6b" by NVIDIA, licensed under CC BY 4.0. Changes: none to the model weights; used for keyword spotting inside our own inference code. |
| FastConformer Hybrid Large Streaming Multi (English) NVIDIA |
CC BY 4.0 | "stt_en_fastconformer_hybrid_large_streaming_multi" by NVIDIA, licensed under CC BY 4.0. Changes: none to the model weights; run in our own streaming inference code. |
| Qwen3-ASR 0.6B and Qwen3-ASR 1.7B Qwen team, Alibaba Cloud |
Apache 2.0 | Copyright Alibaba Cloud. Licensed under the Apache License, Version 2.0. Used unmodified to double-check words before they are confirmed. |
| Whisper small.en OpenAI; converted weights by SYSTRAN |
MIT | Copyright (c) 2022 OpenAI. Used unmodified (in a converted format for faster inference) to double-check words. |
| ECAPA-TDNN speaker embeddings (spkrec-ecapa-voxceleb) SpeechBrain |
Apache 2.0 | Used unmodified to check that the main speaker is the one talking at live events. |
Software
| Project | Licence | Used for |
|---|---|---|
| NVIDIA NeMo | Apache 2.0 | Loading and running speech models. Copyright NVIDIA Corporation & affiliates. |
| vLLM | Apache 2.0 | Fast model serving. Copyright the vLLM project contributors. |
| qwen-asr | Apache 2.0 | Running Qwen3-ASR models. |
| faster-whisper | MIT | Running Whisper models. Copyright (c) 2023 SYSTRAN. |
| CTranslate2 | MIT | Inference engine for converted models. |
| SpeechBrain | Apache 2.0 | Speaker embeddings. |
| PyTorch and torchaudio | BSD-style | Running neural networks. |
| Transformers | Apache 2.0 | Model components. |
| PyTorch Lightning | Apache 2.0 | Model loading (via NeMo). |
| SentencePiece | Apache 2.0 | Tokenisation. |
| NumPy, SciPy | BSD-3-Clause | Audio and numeric processing. |
| librosa | ISC | Audio processing. |
| python-soundfile | BSD-3-Clause | Reading audio files. |
| FFmpeg | LGPL 2.1+ / GPL 2+ (as built) | Audio decoding and conversion. Run as a separate, unmodified program. |
| websockets | BSD-3-Clause | Streaming connections. |
| aiohttp | Apache 2.0 | Web server. |
| cryptography | Apache 2.0 or BSD | Signing and security. |
| Caddy | Apache 2.0 | HTTPS and web serving. |
| Inter, Archivo Black, IBM Plex Mono | SIL Open Font License 1.1 | Fonts on this site. |
Licence texts
- Creative Commons Attribution 4.0 International: creativecommons.org/licenses/by/4.0/legalcode
- Apache License, Version 2.0: apache.org/licenses/LICENSE-2.0
- MIT License: opensource.org/license/mit
- BSD 3-Clause License: opensource.org/license/bsd-3-clause
- SIL Open Font License 1.1: openfontlicense.org
The models and software above are provided by their authors "as is", without warranty. Their authors do not endorse Saidwire. Spot a missing credit? Email support@saidwire.com and we'll fix it.