This is already handled automatically when using models like GPT-4o-Transcribe and Deepgram Nova. They convert spoken numbers into numeric values based on context. For example, “one more thing” stays as words, but “one thousand” becomes 1,000.
In the next update, this will also be improved for the local model (Parakeet). It will convert more consistently, including things like “five dollars” → $5, and spoken dates like “march fifth” → March 5. For example, “twenty dollars on march fifth” would become $20 on March 5.
Log in to comment and vote
Comments1
Utter
Mar 27
Great suggestion, thanks for sharing this.
This is already handled automatically when using models like GPT-4o-Transcribe and Deepgram Nova. They convert spoken numbers into numeric values based on context. For example, “one more thing” stays as words, but “one thousand” becomes 1,000.
In the next update, this will also be improved for the local model (Parakeet). It will convert more consistently, including things like “five dollars” → $5, and spoken dates like “march fifth” → March 5. For example, “twenty dollars on march fifth” would become $20 on March 5.
Appreciate you calling this out!