Chorus v1: Overlapping Speech Transcription, Solved
🎙️ Enterprise Voice AI Services (ASR, TTS, Agents): Trelis.com/voice-ai-services
Video Links:
https://router.trelis.com/transcribe/...
https://huggingface.co/Trelis/Chorus-v1
https://huggingface.co/spaces/Trelis/...
https://huggingface.co/Trelis/Chorus-...
Trelis Links:
Apply for a Trelis Grant: https://trelis.com/trelis-ai-grants/
Join the Trelis Team: https://trelis.com/jobs
Get Trelis AI Tutorials by Email
Subscribe on Substack: https://trelis.substack.com
TIMESTAMPS:
0:00 Introduction to Trelis Chorus V1 model
0:06 Model transcribes overlapping speakers using separate tokens
1:10 Release announcement: router.Trelis.com with free monthly transcriptions
1:25 Available on Hugging Face Space and as GGML file
2:14 Performance: 9% character error rate on AMI test set
2:28 Current version supports two speakers, four-speaker version in development
3:18 Novel training approach: separate forward passes per speaker
3:32 Model card and Q&A available in comments
3:45 Enterprise fine-tuning services for custom datasets
3:52 Use cases: meetings with crosstalk and parliamentary speech
4:00 Closing remarks and invitation for questions