| Cue v5
|
| Copyright 2026 IOTEverythin. Licensed under the Apache License, Version 2.0 (see LICENSE).
|
|
|
| Third-party components and data
|
|
|
| encoder/ The encoder of OpenAI Whisper small (openai/whisper-small), Apache License 2.0.
|
| Modification: only the encoder is kept and its positional-embedding table is cut to the
|
| first 100 positions (2 seconds of audio). All other weights are unchanged.
|
|
|
| cue.safetensors, config.json
|
| Cue head, IOTEverythin. Trained (fine-tuned from Cue v4) on:
|
| - synthetic Indian English calls generated with Chatterbox TTS (Resemble AI, MIT) and, for a
|
| subset of caller lines, Dia-1.6B (Nari Labs, Apache-2.0), with reference voices from GLOBE
|
| (CC0, Mozilla Common Voice) and Svarah (AI4Bharat, CC-BY-4.0);
|
| - the AMI Meeting Corpus (University of Edinburgh, IDIAP, TNO and the AMI Consortium),
|
| Creative Commons Attribution 4.0: https:
|
| - the ICSI Meeting Corpus (International Computer Science Institute), Creative Commons
|
| Attribution 4.0: https:
|
| - background noise from MUSAN (OpenSLR 17, CC-BY-4.0), in training augmentation only.
|
| No audio from these corpora is included in this repository.
|
|
|
