AIMarkTechPostabout 5 hours ago
Open ASR Models Compared: WER, Languages, Latency
10 min read
Cohere Transcribe 2B leads the Hugging Face Open ASR Leaderboard at 5.42% WER, but IBM Granite Speech 4.1 2B (5.33%) and ARK-ASR-3B (5.04%) have since posted lower numbers. Leaderboard scores are not directly comparable due to differing test sets. Throughput varies by over an order of magnitude, making cost per audio-hour the deciding factor for most users.
Level
Hype check
Tap to vote and see what everyone thinks.
Summary by ByteBrief
