Use of Non-textual Training Data for Singapore's Common Languages in National Multimodal Large Language Model Programme

Dennis Tan inquired whether the National Multimodal Large Language Model Programme would use non-textual training data for Singapore’s common languages like Singlish, Malay, Tamil, and Chinese dialects to improve voice recognition and accessibility.
Mrs Josephine Teo responded that the SEA-LION model is already trained on datasets including Singlish, Chinese, Malay, and Tamil, with plans to incorporate speech data containing tone and pitch to enhance it.
The programme will first evaluate performance using non-textual data in standard and colloquial English before expanding to other languages, while monitoring technological developments and adapting plans accordingly.

Dennis Tan inquired whether the National Multimodal Large Language Model Programme would use non-textual training data for Singapore’s common languages like Singlish, Malay, Tamil, and Chinese dialects to improve voice recognition and accessibility.
Mrs Josephine Teo responded that the SEA-LION model is already trained on datasets including Singlish, Chinese, Malay, and Tamil, with plans to incorporate speech data containing tone and pitch to enhance it.
The programme will first evaluate performance using non-textual data in standard and colloquial English before expanding to other languages, while monitoring technological developments and adapting plans accordingly.

Dennis Tan inquired whether the National Multimodal Large Language Model Programme would use non-textual training data for Singapore’s common languages like Singlish, Malay, Tamil, and Chinese dialects to improve voice recognition and accessibility.
Mrs Josephine Teo responded that the SEA-LION model is already trained on datasets including Singlish, Chinese, Malay, and Tamil, with plans to incorporate speech data containing tone and pitch to enhance it.
The programme will first evaluate performance using non-textual data in standard and colloquial English before expanding to other languages, while monitoring technological developments and adapting plans accordingly.

Dennis Tan inquired whether the National Multimodal Large Language Model Programme would use non-textual training data for Singapore’s common languages like Singlish, Malay, Tamil, and Chinese dialects to improve voice recognition and accessibility.
Mrs Josephine Teo responded that the SEA-LION model is already trained on datasets including Singlish, Chinese, Malay, and Tamil, with plans to incorporate speech data containing tone and pitch to enhance it.
The programme will first evaluate performance using non-textual data in standard and colloquial English before expanding to other languages, while monitoring technological developments and adapting plans accordingly.

Summary