Cohere shows why regional speech models still matter
Cohere released a 2B speech-recognition model designed around Arabic dialects, Arabic-English code-switching and domain vocabulary.
Why it matters: A language-specific model can outperform a generic global system when dialect, code-switching and specialist vocabulary dominate the errors.