Avalon: A speech recognition model optimized for human-computer interaction
2 days ago
- #AI
- #speech-recognition
- #transcription
- Avalon is a new speech recognition model optimized for human-computer interaction.
- It outperforms Whisper Large v3 and ElevenLabs Scribe on most OpenASR benchmarks.
- Avalon excels in domains like software and coding, with improved transcription accuracy.
- A new benchmark called AISpeak was created to test Avalon's performance with AI jargon and domain-specific terms.
- Avalon achieved 97.4% accuracy on AISpeak-10, significantly higher than competitors.
- The model was built to address real-world usage, focusing on writing prompts, messages, and emails.
- Avalon does not use user audio or transcripts for training unless explicitly opted in.
- The model is available in Aqua for English, with multilingual versions coming soon.