Cactus-Compute/whistle — new model trending #30 on Hugging Face
Cactus Compute released Whistle, a 16.9 MB on-device speech-to-text model for seven languages without a GPU.
Cactus Compute released Whistle, a 16.9 MB on-device speech-to-text model for phones, wearables, robots, and microcontrollers. It runs on the same CPU engine and .cact container as Needle, with no GPU or extra runtime. It transcribes up to 30 seconds of 16 kHz audio in English, German, French, Spanish, Italian, Dutch, and Polish, and also produces word timestamps and speech embeddings. The card reports word-error-rate results over 86,174 utterances versus published Whisper and Moonshine figures, and says no test audio was in training data.
- Whistle is a single 16.9 MB file that runs on CPU only.
- It transcribes seven languages and emits word timestamps and embeddings.
- It reuses Needle's .cact container, quantisation, and KV cache.
- WER was measured on 86,174 utterances against published Whisper and Moonshine scores.
- Decoder depth is selectable at load time from two layers upward.
- Vendors
- Cactus Compute
- Products
- Needle
- Organizations
- Cactus Compute
From the original article
4 min read at huggingface.coautomatic-speech-recognition · 2,249 downloads · 138 likes · license apache-2. 0
By Cactus-Compute · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.