Frontier speech demos — recording sources and modifications Office, living room and kitchen (10 s each) DEMAND by Joachim Thiemann, Nobutaka Ito and Emmanuel Vincent. https://doi.org/10.5281/zenodo.1227121 Creative Commons Attribution-ShareAlike 3.0 Unported, as specified in the authors' description. https://creativecommons.org/licenses/by-sa/3.0/ Each excerpt is shared under that same license. The files cover seconds 5–15 of OOFFICE/ch08.wav, DLIVING/ch03.wav and DKITCHEN/ch08.wav from the 16 kHz archives. These are listening examples. The reported silence-gate benchmark used separate evaluation audio. Overlapping meeting AMI Meeting Corpus, AMI Consortium. CC BY 4.0. https://groups.inf.ed.ac.uk/ami/download/ Table microphone, meeting EN2002a, 697.23–703.29 s, converted to 16 kHz mono. Sintra in heavy rain FLEURS by Google, English test example en_us-1950. CC BY 4.0. https://huggingface.co/datasets/google/fleurs Mixed at 0 dB signal-to-noise ratio with ESC-10 rain 1-17367-A-10, derived from “unlocki door to mod rain.wav” by cognito perceptu (CC0). https://github.com/karolpiczak/ESC-50/blob/master/LICENSE Both paired evaluation recordings use saved raw recognitions from Orukeet v0.1.0 and OrukQwen 2 (revision 14f7a6f3fa76183b94a5771b03f76a0b31a6e5ef). https://huggingface.co/oruk/orukqwen-asr-1.7b-v2 The frozen audio hashes, reference and original recognitions are in recognitions.json. British English telephone recording MINDS-14, PolyAI, en-GB, PAY_BILL/60268fd7c7abb7f832138d5d.wav. CC BY 4.0. https://huggingface.co/datasets/PolyAI/minds14 Converted to 16 kHz mono PCM. Nigerian English recording Copyright 2018, 2019 Google, Inc. Nigerian English speech corpus, ngf_00295_00053583157.wav. https://www.openslr.org/70/ Converted to mono 16 kHz PCM. Excerpt shared under the source license, CC BY-SA 4.0. https://creativecommons.org/licenses/by-sa/4.0/ India and South Asia recording Mozilla Common Voice contributors, common_voice_en_19451818.mp3. CC0. Accent is the self-reported regional category. Source subset by Nathan Roll. https://huggingface.co/datasets/NathanRoll/CommonVoice_accent_stratified Converted to 16 kHz mono PCM. Room-tone and accent replays orukeet-replays.json contains actual Orukeet v0.1.0 outputs on these exact WAVs. Times were measured on an Apple Silicon CPU using sherpa-onnx 2.0.2, greedy decoding, three CPU threads and already-loaded weights. Single-clip observations. The silence references are no-speech annotations. The accent references are the published source transcripts. The price of groceries Open Yap 1K, The Agentic Data Company. CC BY 4.0. See /samples/conversations/ATTRIBUTION.txt. https://huggingface.co/datasets/TheAgenticDataCompany/open-yap-1k Saved Whisper tiny.en and Oruk Resonance outputs are in /examples/whisper-grocery-result.json. CC BY 4.0: https://creativecommons.org/licenses/by/4.0/ CC0: https://creativecommons.org/publicdomain/zero/1.0/