A Coding Guide to Cactus Whistle: Does a 16.9 MB Speech Model Hold Up Against Whisper Base? Word Error Rate, Speed, Streaming, Timestamps, Keywords, and Speech-to-Tool Calls on a CPU

Cactus Compute's Whistle, a 16.9 MB CPU-based speech-to-text model, was benchmarked against OpenAI's Whisper models on multiple metrics including word error rate and speed. Whistle showed advantages over Whisper base on some tests and maintained competitive performance with faster processing on an Apple M4 CPU. The evaluation also covered streaming latency, timestamps, keyword biasing, and multilingual transcription.