Whisper alternative aimed at live voice use cases

[petewarden]

I'm a long-time fan and user of Whisper, but it's been a struggle using it for real-time applications, so I wanted to share some new models I've just open sourced that support streaming. We've managed to get 6.65% WER on HuggingFace's OpenASR Leaderboard with a 245 million parameter model, compared to Whisper Large v3's 7.44%.

I've learned a lot from lurking here, and I'd love to hear thoughts from the community on if this is an interesting direction?