RealSmile
Mini toolsChorus v1
Chorus-v1 is a multi-speaker speech recognition model built for two-person conversations such as calls, interviews, and podcast segments. Trelis says it is a fine-tune of openai/whisper-large-v3-turbo that takes up to 30 seconds of mono audio and, using <|speaker1|> and <|speaker2|> decoder tokens, returns separate timestamped transcripts for each speaker in two passes. The model is English-only, supports overlapping speech up to about 80% overlap in its AMI evaluation, and is released under Apache 2.0 with open weights.
New
Multimodal
Gen 3
Overview
Chorus-v1 is Trelis’s open-weight two-speaker ASR model for short mono audio clips with overlapping speech. It is a fine-tune of Whisper Large v3 Turbo that outputs a separate timestamped transcript for each of two speakers, without requiring an upstream diarization step.
About Trelis Research
Trelis Research is an AI research company building open-weights voice AI models, ML fine-tuning tools, and hosted GPU inference. Its flagship model Chorus-v1 is a multi-speaker transcription fine-tune of Whisper that separates overlapping speakers. It also offers custom enterprise voice AI (ASR & TTS) services and ML tutorials for developers.
View Company ProfileTools using Chorus v1
No tools found for this model yet.
Last updated: April 21, 2026
