Whistle
Whistle is a speech-to-text model (16.9 MB file (parameters not stated)) from Cactus Compute, released under the Apache-2.0 license. The reviewed source lists 7 languages (7 explicitly enumerated in our record), and the model is designed for native streaming. It is a specialist rather than a general pick: its best placement in the common setups we test is #6 of 26, for live captions and voice input. Use the matcher with your own language and hardware to see where it fits.
A single 16.9 MB streaming speech-to-text model for phones, wearables, robots, smart-home devices and microcontrollers. Seven listed languages, word-level timestamps, CPU-only with no GPU or dependencies, Apache-2.0. Ratings are provisional until independent benchmarks are available.
Key facts
| Provider | Cactus Compute |
|---|---|
| Model family | Whistle |
| Parameters | 16.9 MB file (parameters not stated) |
| License | Apache-2.0 · Permissive — license text ↗ |
| Language coverage | 7 listed languages · 7 enumerated in our record |
| Streaming | Native streaming architecture |
| Timestamps | Yes — in the model or its runtime family |
| Speaker diarization | No — needs a separate component |
| Speech translation | No |
| Documented runtimes | cactus-needle (Python, CLI, C API, WASI) — how to run it |
| Primary source | huggingface.co/Cactus-Compute/whistle ↗ |
| Rating status | Provisional (added November 3, 2026) |
| Last reviewed |
Hardware and workload fit
These are Speech Model Match's conservative 1–5 editorial ratings, relative to the other 19 reviewed models. They come from model size, official memory guidance and documented runtimes — not from a universal benchmark. How the ratings work.
| Dimension | Rating | Scale |
|---|---|---|
| CPU-only fit | 5/5 | |
| Apple Silicon fit | 4/5 | |
| NVIDIA GPU fit | 2/5 | |
| Speed | 5/5 | |
| Accuracy (relative) | 2/5 | |
| Low memory use | 5/5 | |
| Long recordings | 2/5 | |
| Live / low latency | 5/5 | |
| Subtitle work | 3/5 | |
| Edge / low-power devices | 5/5 |
Where it is strong
- Native streaming architecture for live, low-latency transcription
- CPU-only fit: rated 5/5
- Apple Silicon fit: rated 4/5
- Speed: rated 5/5
- Low memory use: rated 5/5
- Live / low latency: rated 5/5
- Edge / low-power devices: rated 5/5
What to check first
- NVIDIA GPU fit: rated 2/5
- Accuracy (relative): rated 2/5
- Long recordings: rated 2/5
- No native speaker diarization — pair it with a separate diarization pipeline if you need speaker labels
How it ranks in common setups
Ranks come from the same deterministic matcher that runs on the homepage, applied to all 26 reviewed models. They are decision guidance, not accuracy claims.
| Setup | Rank | Fit score | Top match |
|---|---|---|---|
| Long recordings on an 8–12 GB NVIDIA GPU, balanced | #26 of 26 | 54/100 · Weak fit | Whisper turbo |
| Live captions or voice input (native streaming + timestamps, speed first) | #6 of 26 | 82/100 · Strong fit | Nemotron 3.5 ASR Streaming 0.6B |
| Subtitles with timing, accuracy first | #25 of 26 | 57/100 · Weak fit | Whisper medium |
| Meetings with speaker labels (diarization + timestamps) | #26 of 26 | 51/100 · Weak fit | VibeVoice-ASR 7B |
| CPU-only machine, low memory use | #6 of 26 | 71/100 · Possible fit | VibeVoice-ASR-BitNet |
| Apple Silicon with 8–16 GB unified memory, long recordings | #18 of 26 | 63/100 · Possible fit | Whisper small |
| Japanese live transcription (native streaming, speed first) | Not ranked (language not listed) | — | Nemotron 3.5 ASR Streaming 0.6B |
Supported languages
Compare coverage across every reviewed model in the language support matrix.
Community activity on Hugging Face
- Downloads, last 30 days
- 90,000
- Likes
- 566
- Repository last updated
- 5 Nov 2026
- Published on the Hub
- 30 Sept 2026
Figures from the Hugging Face Hub API, 11 Nov 2026. Source: huggingface.co/Cactus-Compute/whistle ↗. Downloads count file downloads, not active users, and do not affect the matcher's scores. Compare all models.
How to run Whistle locally
Whistle runs through Cactus's needle package on the CPU, with no GPU and no other dependencies. Runtimes below are the ones the official model card or repository documents for this model (checked November 2, 2026).
- cactus-needle (Python, CLI, C API, WASI) ↗Official package · CPU only; live streaming mode
Install command shown in the official huggingface.co page ↗:
pip install cactus-needle
Audio input: Live audio streams of any length; each word comes with start, end and probability.
Official hardware guidance: A single 16.9 MB file that runs on CPU with no GPU and no dependencies. Source ↗
Whatever runtime you choose, measure speed and accuracy on your own audio with the benchmark tool; results differ a lot between runtimes, precision and hardware.
Before you deploy Whistle
A printable checklist built from the facts on this page. Tick it off before putting the model into production.
- Language. Confirm your language is in the 7-language list on this page, and test any accents or code-switching you expect.
- License. Apache-2.0: Give recipients a copy of the license, keep copyright and attribution notices, carry forward any NOTICE file and mark files you changed. The patent license ends if you start patent litigation over the work. License guide.
- Install. Start from the official command:
pip install cactus-needle(full run notes). - Audio input. Live audio streams of any length; each word comes with start, end and probability.
- Hardware. A single 16.9 MB file that runs on CPU with no GPU and no dependencies.
- Features you rely on. For Whistle, native streaming is documented; timestamps are documented; speaker labels need a separate diarization component. Test each one you need with real audio.
- Measure. Run 10–20 representative clips, score them with the WER calculator and record speed in the benchmark workspace before you commit.
- Provisional ratings. This model was added on November 3, 2026; treat its hardware and workload ratings as estimates and weight your own measurements higher.
Alternatives and related models
Similar fit profile: Moonshine Tiny Streaming (English) (Moonshine AI, 34M), VibeVoice-ASR-BitNet (Microsoft, 0.3B quantized), Whisper tiny (OpenAI, 39M). Similarity is computed from the ten fit ratings plus streaming and diarization support.
| Fact | Whistle | Moonshine Tiny Streaming (English) | VibeVoice-ASR-BitNet |
|---|---|---|---|
| Provider · size | Cactus Compute · 16.9 MB file (parameters not stated) | Moonshine AI · 34M | Microsoft · 0.3B quantized |
| License | Apache-2.0 Permissive | MIT Permissive | MIT Permissive |
| Languages | 7 listed languages | English | 7 listed languages |
| Streaming | Native | Native | Via runtime |
| Timestamps | Yes | Limited | Limited |
| Speaker diarization | Separate pipeline | Separate pipeline | Separate pipeline |
| CPU · Apple · NVIDIA fit | 5/5 · 4/5 · 2/5 | 5/5 · 5/5 · 4/5 | 5/5 · 4/5 · 3/5 |
| Live · long recordings | 5/5 · 2/5 | 5/5 · 3/5 | 4/5 · 4/5 |
| Main runtime | cactus-needle (Python, CLI, C API, WASI) | moonshine-voice (Python) | VibeASR.cpp |
Head-to-head pages: Moonshine Tiny Streaming (English) vs Whistle
Check Whistle for your setup
Pick your language, hardware and workload to see this model's fit score, the reasons behind it and where it ranks against all reviewed models. Everything runs in your browser.
Common questions
Can I use Whistle commercially?
Permissive license; commercial use is allowed under the listed terms. The listed license is Apache-2.0; read the current license text ↗ before shipping a product, because terms can change. The license guide compares the obligations of all 26 reviewed models.
Does Whistle support real-time streaming?
Yes. The reviewed source describes a native streaming architecture, so it is a candidate for live captions, dictation or voice agents. Measure end-to-end latency on your own hardware.
Which languages does Whistle transcribe?
7 listed languages. The full list of 7 explicitly enumerated languages is shown above.
Can Whistle tell speakers apart?
Not natively. Combine it with a dedicated diarization component and evaluate that component separately; the meetings guide covers this workflow.
Primary sources
Reviewed against primary sources on November 3, 2026. Model behavior, runtime support, language coverage and licenses can change; re-check the linked primary source before production use. See the full source directory.