arxiv:2410.15608

Moonshine: Speech Recognition for Live Transcription and Voice Commands

Published on Oct 21, 2024

Authors:

Abstract

Moonshine, an encoder-decoder transformer architecture for speech recognition, uses Rotary Position Embedding, reducing compute requirements without decreasing accuracy.

Generated by Qwen/Qwen2.5-Coder-32B-Instruct

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.

View arXiv page View PDF GitHub 11.3k auto Add to collection

Community

ali073ml

4 days ago

Hello Moonshine AI Team,
My name is Nomi (nzg73). I have been following your work on Hugging Face and am really impressed by the open-source Moonshine ASR models you have released across multiple languages.

Currently, there is a lack of accurate open-source ASR models for Pakistani Urdu. Most existing models on Hugging Face struggle heavily with Urdu pronunciation, phonetics, and vocabulary, often generating jumbled or incorrect transcriptions. Since you are experts in building efficient speech-to-text models, I would love to request your support in creating a Moonshine base model for the Urdu language.

To help make this happen, I am ready to actively contribute:
Dataset Contribution: I have a large dataset of over 50,000 Urdu samples that I am willing to share with you for training.
Linguistic Support: As a native speaker, I can fully assist with accurate Urdu vocabulary, script alignment, and pronunciation guidelines.
While I am unfamiliar with the exact architecture requirements for training an ASR model from scratch, I would appreciate any guidance on how ASR models are built and how the audio/text data should be formatted for your pipeline. If you can help initialize or train a base Urdu model, I can take it from there to further fine-tune it on additional data and gladly share the updated checkpoints back with you and the community.
Please let me know how we can collaborate on this. Looking forward to your response!
Best regards,
Nomi (nzg73)