Moonshine: Speech Recognition for Live Transcription and Voice Commands

TL;DR

Moonshine is a new speech recognition model based on transformer architecture, optimized for live transcription and voice commands, offering significant efficiency improvements without sacrificing accuracy.

Contribution

Introduces Moonshine, a transformer-based speech recognition model using Rotary Position Embedding and no zero-padding, enhancing efficiency for real-time applications.

Findings

5x reduction in compute for transcribing 10-second segments

No increase in word error rates compared to Whisper tiny-en

Effective for resource-constrained, real-time speech recognition

Abstract

This paper introduces Moonshine, a family of speech recognition models optimized for live transcription and voice command processing. Moonshine is based on an encoder-decoder transformer architecture and employs Rotary Position Embedding (RoPE) instead of traditional absolute position embeddings. The model is trained on speech segments of various lengths, but without using zero-padding, leading to greater efficiency for the encoder during inference time. When benchmarked against OpenAI's Whisper tiny-en, Moonshine Tiny demonstrates a 5x reduction in compute requirements for transcribing a 10-second speech segment while incurring no increase in word error rates across standard evaluation datasets. These results highlight Moonshine's potential for real-time and resource-constrained applications.