Case Study: 10K hours of Bengali Speech at Scale
How Poseidon built 10,000 hours of native-speaker Bengali speech data, and how fine-tuning Whisper on it cut character error rate by ~32%.
Case StudiesRashmi NagpalNick Ippolito
More from the Poseidon team
11
ResearchBuilding Better Speech Data for Indic Languages
Yash VyasKrishna Rupaakula
ProductOpen-Sourcing SONAR: Poseidon's Multilingual ASR Evaluation Toolkit
Rashmi NagpalAndy Bo Wu
ResearchSONAR: Evaluating Voice AI Beyond English
Rashmi NagpalSarick ShahLJW
ResearchThe Complexity of Multi-Speaker Audio Evaluation
Krishna RupaakulaSebastian LiuLJW
ResearchWe Tested 3 Frontier Models on Bengali Linguistics: They Rarely Agreed
Sebastian LiuKrishna RupaakulaLJW
CompanySovereign AI Needs Verifiable Trust
Poseidon
ProductIntroducing the Poseidon Voice AI Dataset
Poseidon
CompanyIntroducing Poseidon Litepaper
Poseidon
ProductLive Now: Poseidon App V1
Poseidon
ResearchWhy Distributed Systems Matter for AI’s Data Bottleneck
Poseidon
CompanyPoseidon Raises $15M Seed Round to Accelerate Physical AI
Poseidon
