Papers

These research papers are a good resource for understanding the architectures and performance strategies behind the models:

• Moonshine: Speech Recognition for Live Transcription and Voice Commands: Describes the first-generation model architecture, which enabled flexible-duration input windows, improving on Whisper's fixed 30 second requirement.
• Flavors of Moonshine: Tiny Specialized ASR Models for Edge Devices: How we improved accuracy for non-English languages by training mono-lingual models.
• Moonshine v2: Ergodic Streaming Encoder ASR for Latency-Critical Speech Applications: Introduces our approach to streaming, and the advantages it offers for live voice applications.