{"apiVersion":"v1","taxonomyVersion":1,"generatedAt":"2026-10-29T16:39:21.449Z","data":{"recordId":"wsa_6p9gfcxrrshe","score":1,"confidence":"high","capped":false,"capReason":null,"sources":[{"n":1,"url":"https://huggingface.co/Cactus-Compute/whistle","publisher":"Hugging Face","authority":"owner-controlled","verification":"verified","lastRetrievedAt":"2026-10-28T00:27:38.878Z","detectedPublishedAt":"2026-09-30T17:13:26.000Z","rights":"original"},{"n":2,"url":"https://cactuscompute.com/blog/whistle","publisher":"Cactus Compute","authority":"owner-controlled","verification":"verified","lastRetrievedAt":"2026-10-28T00:27:38.878Z","detectedPublishedAt":"2026-10-02T00:00:00.000Z","rights":"original"},{"n":3,"url":"https://cactuscompute.com/blog/finetuning-whistle","publisher":"Cactus Compute","authority":"owner-controlled","verification":"verified","lastRetrievedAt":"2026-10-18T16:58:38.336Z","detectedPublishedAt":"2026-10-18T00:00:00.000Z","rights":"original"}],"claims":[{"id":"a8fe4b0a-1571-4c1c-b991-601d7a2ea0f5","kind":"capability","statement":"Fine-tuning adapts Whistle to speakers and vocabulary or teaches it a new language.","citations":[{"source":3,"supportProbability":0.95,"supports":true,"quote":{"text":"needle whistle finetune adapts it to your speakers and vocabulary, or teaches it a new language, on your own machine.","rights":"original"}}]},{"id":"afe1b988-eca1-4fe0-869a-32df7faa0a93","kind":"deployment","statement":"Fine-tuning runs on the user's own machine.","citations":[{"source":3,"supportProbability":0.93,"supports":true,"quote":null}]},{"id":"8338c9f5-4cae-4ab5-917a-2f2c993d186f","kind":"capability","statement":"The build command merges the adapter and writes a .cact file that carries the languages and tokens.","citations":[{"source":3,"supportProbability":0.98,"supports":true,"quote":{"text":"build merges the adapter and writes a .cact that carries your languages and tokens.","rights":"original"}}]},{"id":"fe9d75db-3a1f-4337-ba8f-e307874b9d64","kind":"capability","statement":"Streaming, keywords and word timestamps work as they do with the shipped model.","citations":[{"source":3,"supportProbability":0.99,"supports":true,"quote":{"text":"Streaming, keywords and word timestamps work as they do with the shipped model.","rights":"original"}}]},{"id":"67edabed-bb18-46e5-8373-2da692209177","kind":"recommendation-relevant","statement":"Brazilian Portuguese fine-tuned with about ten hours of FLEURS data reached 30.5% word error rate on the FLEURS test set.","citations":[{"source":3,"supportProbability":0.97,"supports":true,"quote":{"text":"Brazilian Portuguese from about ten hours of FLEURS, at --lr 1e-2 with the other defaults, went from 106.2% to 30.5% word error rate on FLEURS test and was detected as Portuguese on every test clip.","rights":"original"}}]},{"id":"14371f1c-b24e-4d90-8746-26acec8a0177","kind":"privacy","statement":"In the browser sandbox, audio never leaves the device.","citations":[{"source":2,"supportProbability":0.84,"supports":true,"quote":{"text":"The first press downloads the 16.9 MB model, and audio never leaves your device.","rights":"original"}}]},{"id":"22f105ba-9e03-4f65-9a26-668c091dcc08","kind":"capability","statement":"Whistle transcribes 16 kHz mono audio in English, German, French, Spanish, Italian, Dutch, and Polish.","citations":[{"source":1,"supportProbability":0.97,"supports":true,"quote":{"text":"- **Transcription**: 16 kHz mono audio of any length, one pass up to 30 seconds and windowed beyond it, in English, German, French, Spanish, Italian, Dutch and Polish.","rights":"original"}}]},{"id":"c4f52ef5-8fb9-4d75-9698-0cd69c93f4f9","kind":"capability","statement":"Whistle provides word start and end times and a probability for every word.","citations":[{"source":1,"supportProbability":0.98,"supports":true,"quote":{"text":"- **Word timestamps**: every word with its start, end and probability, aligned from the decoder's own attention, so an app can highlight, seek or cut on a word.","rights":"original"}}]},{"id":"b6ae1a7c-70ea-4a8f-9d90-d3db84136350","kind":"capability","statement":"Whistle provides speech embeddings with one row per 80 ms frame.","citations":[{"source":1,"supportProbability":0.99,"supports":true,"quote":{"text":"- **Speech embedding**: the encoder output, one row per 80 ms frame, for matching and retrieval without decoding a transcript at all.","rights":"original"}}]},{"id":"fca865fc-e924-4aa5-aaf6-dfb1d088631b","kind":"availability","statement":"The repository contains the downloadable whistle.cact model file.","citations":[{"source":1,"supportProbability":0.94,"supports":true,"quote":null}]},{"id":"555064ac-a8ed-4693-bdc8-29c71a4f14ca","kind":"deployment","statement":"Whistle runs on-device on a CPU without a GPU.","citations":[{"source":1,"supportProbability":0.96,"supports":true,"quote":{"text":"The whole model is a single 16.9 MB file, and it runs on the same CPU engine as [Needle](https://huggingface.co/Cactus-Compute/needle3), from the same container and the same quantisation, with no dependencies and no GPU.","rights":"original"}}]},{"id":"b90e0a8d-98f9-41de-878a-85a915c38129","kind":"recommendation-relevant","statement":"The complete model is a single 16.9 MB file.","citations":[{"source":1,"supportProbability":0.95,"supports":true,"quote":{"text":"The whole model is a single 16.9 MB file, and it runs on the same CPU engine as [Needle](https://huggingface.co/Cactus-Compute/needle3), from the same container and the same quantisation, with no dependencies and no GPU.","rights":"original"}}]}]},"nextCursor":null,"license":{"catalog":{"name":"CC BY 4.0","spdx":"CC-BY-4.0","url":"https://creativecommons.org/licenses/by/4.0/","attribution":"wowsuchai","attributionUrl":"https://wowsuchai.withglint.app","covers":["catalog metadata","classifications","generated summaries","feed event metadata"]},"excluded":{"covers":["source quotations","linked source content"],"terms":"Not licensed by wowsuchai. Source quotations and linked source content keep their original rights."}}}